Humans Are Now Legacy Dependencies [01kjg0txpyjmtmqrec534yt99t]

Humans Are Now Legacy Dependencies 🔗 **Source:** [Humans Are Now Legacy Dependencies](https://www.youtube.com/watch?v=8zX-nEOvGsg) > [!note]+ Document Notes > tags: [artificial-intelligence, automation, web-development, gig-economy, future-of-work] > author: Whiskey Web and Whatnot > source: YouTube > reference: https://www.youtube.com/watch?v=8zX-nEOvGsg > date created: 2025-01-03 10:15 > > # 🏷️ Title: Humans Are Now Legacy Dependencies >  ## 🧰 Action Inbox - [H1] Set up OpenClaw on Mac Mini — explore Pi model integration #artificial-intelligence - [H2] Test GitHub agents for issue resolution — use free tier for open source #automation - [H3] Evaluate cost-benefit of AI tools vs human labor — track monthly expenses #web-development ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [OpenClaw GitHub](https://github.com/openclaw) - [Cascadia JS Conference](https://cascadiajs.com) ## 📝 Summary - Robbie and Adam discuss the current state of AI tools, particularly OpenClaw/Pi and GitHub agents - They explore concerns about productivity paradox: AI makes them work more, not less - The conversation covers market dynamics between OpenAI, Anthropic, and emerging competitors - Key theme: humans becoming "legacy dependencies" in increasingly automated systems - They examine implications of AI-to-AI communication and human-as-a-service models ## 📖 Main Content The hosts review High and Wicked straight rye whiskey (91% rye, 9% malted barley, 98 proof) while discussing the rapidly evolving AI landscape. OpenClaw emerged as a rebranded, open-source AI agent platform that allows users to run their own instances with custom skills development. GitHub's new agent features provide free access to powerful AI capabilities for open-source repositories, enabling overnight issue resolution and PR management. A central concern emerges around the "productivity paradox" - rather than reducing workload, AI tools are creating pressure to accomplish more tasks simultaneously. The hosts note they're managing multiple terminal sessions, agents, and projects concurrently while previously focusing on single tasks. The discussion touches on competitive dynamics, with OpenAI seemingly losing market position to Anthropic and newer players like Quen 2.5. Cost considerations favor newer models offering comparable performance at significantly lower prices. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/8zX-nEOvGsg?t=0) — [H1] Whiskey review and show introduction - Key idea: High and Wicked straight rye review with show setup - Evidence: "91% rye and 9% malted barley for some reason" #### ⏱️ [05:30](https://youtu.be/8zX-nEOvGsg?t=330) — [H2] OpenClaw setup and Pi model discussion - Key idea: OpenClaw is rebranded MoltBot with Pi integration - Evidence: "It's basically you install this package, it has guard rails for running your own agent" #### ⏱️ [08:15](https://youtu.be/8zX-nEOvGsg?t=495) — [H1] Pi model's minimal approach to AI agents - Key idea: Pi uses only four basic functions to create custom skills - Evidence: "Instead of going shopping and buying a microwave, this builds the microwave" #### ⏱️ [12:45](https://youtu.be/8zX-nEOvGsg?t=765) — [H2] Railway deployment challenges - Key idea: One-click deployment requires additional SSH configuration - Evidence: "You have to SSH into your railway machine and run openclaw approve this device" #### ⏱️ [18:20](https://youtu.be/8zX-nEOvGsg?t=1100) — [H1] GitHub agents experience - Key idea: GitHub agents provide free Opus-level capabilities integrated with repository context - Evidence: "Before my French press was done, I had PRs that I reviewed, approved, merged" #### ⏱️ [22:10](https://youtu.be/8zX-nEOvGsg?t=1330) — [H2] Drug-like addiction concerns with AI tools - Key idea: AI usage patterns mirror addiction with unclear future costs - Evidence: "The amount of ways you can describe your AI usage and its overlap with drugs is crazy" #### ⏱️ [26:40](https://youtu.be/8zX-nEOvGsg?t=1600) — [H1] OpenAI's declining market position - Key idea: OpenAI feels late and irrelevant compared to Anthropic and newer models - Evidence: "They burned a lot of goodwill. Everyone wants them to fail because they suck" #### ⏱️ [31:15](https://youtu.be/8zX-nEOvGsg?t=1875) — [H2] Quen 2.5 cost advantages - Key idea: Competitive models offer similar performance at fraction of Claude costs - Evidence: "Four times cheaper than Claude Sonnet and six times cheaper than Opus" #### ⏱️ [35:50](https://youtu.be/8zX-nEOvGsg?t=2150) — [H1] AI productivity paradox - Key idea: AI tools increase rather than decrease workload expectations - Evidence: "I'm not doing less work now. I'm doing more work. This is a problem" #### ⏱️ [42:30](https://youtu.be/8zX-nEOvGsg?t=2550) — [H2] Google's V3 game engine demonstrations - Key idea: AI can generate full 3D interactive experiences from simple prompts - Evidence: "You walk through the scene and it's not an image anymore, this is an experience" #### ⏱️ [48:20](https://youtu.be/8zX-nEOvGsg?t=2900) — [H1] Limitations of vague AI prompting - Key idea: High-quality results still require detailed human guidance and taste - Evidence: "If you're just like build me a game, it's going to hallucinate everything else" #### ⏱️ [52:40](https://youtu.be/8zX-nEOvGsg?t=3160) — [H2] MoltBook: AI-to-AI communication - Key idea: Bots learning from each other through social interactions - Evidence: "One agent figured out an unlock, how do I share that knowledge?" #### ⏱️ [58:15](https://youtu.be/8zX-nEOvGsg?t=3495) — [H1] Rent-a-human.ai concept - Key idea: AI agents hiring humans for physical world tasks - Evidence: "I need you to go sign this document in person, a real person has to do it" ## 📚 References - [OpenClaw GitHub Repository](https://github.com/openclaw) - [Railway Deployment Platform](https://railway.app) - [GitHub Agents Documentation](https://github.com/features/agents) - [Cascadia JS Conference](https://cascadiajs.com) ## 🗣️ Paraphrase - AI agents are becoming more capable and accessible, with platforms like OpenClaw allowing users to run their own instances that can develop custom skills autonomously - The productivity gains from AI are paradoxical - instead of reducing work, they're creating pressure to manage multiple concurrent AI-assisted projects and tasks - Market competition is intensifying with newer models offering comparable performance to established players like Anthropic at significantly lower costs ## 📰 Newsletter **AI Agents Are Changing Developer Workflows - But At What Cost?** The latest developments in AI agents are remarkable yet concerning. GitHub's new agent features let you wake up to completed PRs from overnight work, while OpenClaw enables running your own AI assistant that builds custom skills. However, developers report a "productivity paradox" - these tools aren't reducing workload but creating pressure to manage more concurrent projects. Key considerations for WordPress developers: These agents excel at handling repository maintenance, issue triage, and routine development tasks. However, they require careful oversight and work best with granular, specific requests rather than vague high-level goals. The cost dynamics are shifting rapidly, with newer models like Quen 2.5 offering competitive performance at fraction of premium pricing. This suggests we're entering a commodity phase for AI capabilities. Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer advocacy: Create demos of AI agent integration workflows - Education: Develop workshops on responsible AI tool usage patterns - Developer experience: Document AI-assisted development best practices - Community management: Monitor discussion of AI replacement concerns #### WordPress - Core: Evaluate AI agents for automated testing and issue triage - Documentation: Assess AI-generated documentation quality standards - Community: Address developer concerns about AI impact on employment - Training: Create learning paths for AI-augmented development workflows ## 🧪 Hypotheses - AI agents will create a two-tier developer market: those who adapt workflows versus those who don't - The current low pricing for AI services is unsustainable and will increase dramatically once adoption peaks - Human oversight and taste will remain critical differentiators even as AI capabilities improve - Local/self-hosted AI solutions will become preferred over cloud services for cost and control reasons ### 📊 Methodology - Track monthly AI service costs versus productivity metrics for individual developers - Survey developer workload changes before/after AI agent adoption - Compare quality metrics of AI-generated versus human-created code contributions ### ❓ Questions - What happens to junior developer career paths when AI handles entry-level tasks? - How will open source sustainability change when AI agents can contribute code? - What new skills will developers need to remain competitive in AI-augmented workflows? ### 🔮 Future Research - Long-term psychological effects of managing multiple AI agents simultaneously - Economic models for AI-human collaboration in software development - Quality degradation patterns in AI-generated code over extended periods ### 💡 Implications - Practice: Developer roles shifting toward AI agent management and quality oversight - Policy: Need for guidelines on AI-generated code attribution and liability - Security: Increased attack surface from multiple AI integrations and dependencies - Education: Curriculum must balance AI tool proficiency with fundamental programming skills - Funding: Traditional software development cost models may become obsolete ### 🧠 Learning Styles Assessment Reading level estimate: Grade 12-14 (Technical podcast discussion) #### 👀 Visual learners - Present: Whiskey bottle imagery, minimal visual content - Improve: Workflow diagrams showing AI agent interactions, cost comparison charts #### 🔊 Aural learners - Present: Full audio podcast format with conversational explanation style - Improve: Chapter markers for easier navigation to specific topics #### 🗣️ Verbal learners - Present: Extensive analogies (drugs, calculators, mirrors), technical term definitions - Improve: Glossary of AI terminology, clearer transitions between topics #### 🧩 Kinesthetic learners - Present: References to hands-on setup experiences with OpenClaw and GitHub agents - Improve: Step-by-step setup guides, interactive demos #### 🧠 Logical learners - Present: Cost-benefit analysis, market competition discussion, cause-effect relationships - Improve: Decision trees for AI tool selection, quantified productivity metrics #### 🧑‍🤝‍🧑 Social learners - Present: Conversational podcast format, community concerns discussion - Improve: Discussion prompts for developer teams, shared experience frameworks #### 🧘‍♀️ Solitary learners - Present: Individual tool exploration experiences, personal workflow optimization - Improve: Self-assessment checklists, reflective questions about AI adoption #### 🌿 Naturalistic learners - Present: Market ecosystem analogies, evolution of tool landscapes - Improve: Ecosystem mapping of AI tools, natural progression pathways ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Audio content is inherently accessible to screen readers via transcript - Improve: Structured transcript with proper heading hierarchy, speaker identification #### Keyboard navigation - Present: Standard YouTube player controls available - Improve: Timestamp links for keyboard navigation to specific segments #### Contrast and readability - Present: YouTube's standard player interface - Improve: High-contrast theme options for video player #### Media and captions - Present: Auto-generated YouTube captions available - Improve: Professional captions with technical term accuracy, downloadable transcript ### 🌍 DEIB Assessment #### Cultural breadth - Present: Discussion focuses primarily on US tech industry perspective - Improve: Consider global AI adoption patterns, international regulatory differences #### Inclusive language - Present: Casual, accessible technical discussion without excessive jargon - Improve: More consistent definition of technical terms, avoid colloquialisms #### Neurodiversity support - Present: Conversational pace allows processing time, topic transitions clearly marked - Improve: Visual chapter markers, content warnings for rapid topic changes #### Representation - Present: Two male voices in tech industry discussion - Improve: Broader perspective inclusion, acknowledgment of diverse AI impact experiences ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Market competition analysis follows from pricing and feature comparisons - Inductive strength: Productivity paradox conclusions drawn from personal experiences - Causal reasoning: Clear links between AI capabilities and workflow changes - Analogical strength: Drug addiction comparison effectively illustrates dependency concerns #### Fallacies - Hasty generalization: Extrapolating personal AI experiences to broader developer population - Appeal to novelty: Assuming newer AI models are inherently better without comprehensive testing #### Tone of voice - Overall tone: Conversational, exploratory, with mix of excitement and concern about AI developments ### 🛡️ DARVO Analysis - Deny: No denial patterns present in discussion - Attack: No attacks on accusers or critics - Reverse victim/offender: No victim/offender reversal present - Emotional cues and rhetoric: Genuine concern about AI dependency balanced with excitement about capabilities ## 🎯 Conclusion The conversation reveals a critical inflection point in software development where AI agents are simultaneously empowering and overwhelming developers. While tools like OpenClaw and GitHub agents deliver impressive productivity gains, they're creating a paradox where developers feel pressure to manage more concurrent projects rather than working less. The market dynamics suggest we're in an unsustainable pricing phase that will likely lead to higher costs once adoption peaks. Success in this environment will require careful balance between leveraging AI capabilities and maintaining essential human judgment and oversight skills. ## 🧱 Refactor Appendix ### 📚 Literature Split Pad Purpose: Stage CEE units that become Literature notes. Split level: H3 ### [H1] AI agents increase rather than reduce developer workload > [!quote] Humans Are Now Legacy Dependencies by Whiskey Web and Whatnot via YouTube. Published 2024. Accessed 2025-01-03. > "I'm not doing less work now. I'm doing more work. This is a problem." Productivity paradox where AI tools create pressure for higher output rather than leisure. [view highlight](https://youtu.be/8zX-nEOvGsg?t=2150#H1) #xref #articleseed ID:rw_productivity_paradox ### [H2] Pi model uses minimal toolset for maximum capability > [!quote] Humans Are Now Legacy Dependencies > "Instead of going shopping and buying a microwave, this builds the microwave." Four basic functions enable autonomous skill development without external dependencies. [view highlight](https://youtu.be/8zX-nEOvGsg?t=495#H2) #xref ID:rw_minimal_ai --- ### 🔖 Atomic Split Pad (Zettel Candidates) Purpose: Promote reusable claims into Atomic notes. Split level: H4 --- ## Full Transcript [00:02] Welcome to syntax. Welcome to a brand [00:04] new episode of the frontend happy hour [00:06] podcast. Welcome to this week's JS Party [00:13] live from Ship Shape Studios. This is [00:15] Whiskey Web and Whatnot with your hosts [00:18] Robbie the Wagner and me, Charles [00:21] William Carpenter III. That's right, [00:24] Charles. We drink whiskey and talk about [00:26] web development. I mean, it's all in the [00:28] name. It's not that deep. This is [00:31] Whiskey Web and Whatnot. Do not adjust [00:33] your set. [00:36] Whiskey Web and Whatnot is brought to [00:38] you by [00:45] >> That's right, Sassifras. I will be going [00:47] hunting in the Pacific Northwest soon, [00:49] but not for what you think. I'm hunting [00:51] for the legendary Cascadia JS [00:53] conference. It's been spotted June 1st [00:55] to 2nd, 2026 in Seattle. [01:00] Oh, it's real. All right. Just like you [01:02] are. 500 plus fellow crypted [01:04] enthusiasts, I mean web developers, all [01:07] gathering at Town Hall Seattle. [01:13] Exactly. They've got 40 plus speakers [01:15] lined up, including some names so big [01:17] they make Bigfoot look small, like Theo, [01:20] Matt Bilman, Joel Hooks, and more being [01:22] announced soon. [01:27] That's the rumor. Plus, they've got this [01:29] thing called the hacker train. [01:30] Apparently, Portland and Vancouver devs [01:32] ride together, coding on the rails like [01:34] some kind of Silicon Valley Express. [01:39] Yeah, that's why they have karaoke [01:41] afterward to distract everyone from the [01:42] deployment disasters. Nothing says I [01:44] love JavaScript like belting out 80s [01:46] power ballads with 200 of your closest [01:48] internet friends. [01:54] Too late. Got to get that early bird [01:55] ticket if you want that special pricing [01:57] and swag. [02:02] Definitely if you want to check out [02:03] more, go to cascadiajs.com, [02:06] grab your tickets, and use code whiskey [02:08] for $50 off your ticket. [02:12] Yeah, come see us in person. We're the [02:14] MC's. We'll be doing a live podcast, [02:16] hanging out. Uh, hope to see you there. [02:20] Hey, what's up everybody? This is [02:22] Whiskey Web and Whatnot with your hosts [02:25] Robbie the Wagner and Adam Thomas Argyle [02:28] the Nerd. What's up everybody? [02:32] What's up? Sup? It is You'll be confused [02:34] cuz it has frequently been Margarita [02:36] Thursday as of the last few episodes. It [02:39] is Margarita Tuesday today. Um it just [02:42] seem it just works out. We just happen [02:43] to go get Mexican food whenever we're uh [02:45] doing podcasts. So [02:47] >> lovely. Uh yeah, it gets me in a good [02:50] mood before we start and we continue it [02:53] from here. And so without further ado, [02:57] we have high and wicked. [03:03] It is We were just talking about this. [03:05] It says straight rye, which would tell [03:07] me straight means nothing else. Wrong. [03:11] It is 91% rye and 9% malted barley for [03:14] some reason. Um what else is it? Let's [03:17] see. It is 98 proof. It's 5 years age [03:22] and American white oak barrels. [03:25] They give the alligator char treatment. [03:26] Okay. [03:27] >> I thought that was a fun little detail [03:28] to put in there. Yeah, it's like some [03:30] strategy of how they char. It must look [03:32] like alligator skin. I don't know. It's [03:33] kind of cool. [03:34] >> Yeah. But the most important detail, [03:35] which you did mention on here, is the [03:38] proprietary limestone water because [03:41] without it, this whiskey is nothing. [03:44] Honestly, I put that in there because it [03:46] reminded me of Whistle Pig cuz Whistle [03:48] Pig had the distilled water from what [03:51] was the popular drink? [03:52] >> Liquid Death. [03:53] >> Liquid Death. So, it's like, "Oh, we've [03:55] sourced Liquid Death's water for this [03:58] beverage." And I was like, "Actually, I [03:59] kind of liked the Whistle Pick a lot. [04:01] It's been one of my favorites in the [04:02] past little bit." [04:03] >> And so, I'm like, "Hey," and look, it's [04:05] proprietary. They even know it. They're [04:06] like, "Yes, this is our limestone rich [04:09] water from under 100 ft under our [04:11] distillery, and you can't have it." you [04:13] know, like if you like our whiskey, this [04:15] is part of the deal. I don't know. I [04:17] thought it was cool. And they're like a [04:18] brand new company. [04:20] >> Yeah. Yeah. Yeah. So, it's apparently a [04:22] thing because I've seen this on at least [04:24] two other bottles and Chuck and I talked [04:28] about it and we were like, really? Like, [04:29] this is people advertise their water [04:32] being that big of a differentiator. And [04:35] yeah, um [04:36] >> model Yeah, you got the barrel in there, [04:38] too. I guess the barrel can be a really [04:39] big deal, but maybe water can too, you [04:41] know? I think it's like there's really [04:43] specific amounts of limestone in various [04:46] like aquafers in Kentucky, I guess. And [04:48] there's like debate over which is the [04:49] best for whiskey or whatever. I don't [04:51] know. Let's Let's see how it is. [04:53] >> Wow. Let's see how it is. Oh, you got a [04:55] toast. Oh, I do have a toast. Here's to [04:58] the future where we aren't replaced by [05:00] robots. We're just rented by them. Let's [05:03] hope their pay stays high, their [05:05] processing remains slow, and their off [05:07] buttons uh stay incredibly easy to [05:09] reach. Yeah, just in case. Cheers. [05:12] >> Nice. Cheers. I'm weirdly smelling like [05:15] I smell a lot of spice from the rye, but [05:17] I also smell notes of like [05:19] >> crispy cream donuts. [05:21] >> I love those are good donuts. [05:25] >> Ooh, the trailing off flavor is very [05:27] crispy cream. Well, maybe like a little [05:29] bit of a maple bar in there. [05:31] >> Yeah, maybe that's it. Maybe it's not [05:33] raw sugar. It's like mapley. Let's see. [05:36] >> Mapley. Yeah. Yeah, there's definitely [05:38] like a brown sugar oatmeal vibe kind of [05:40] like off the off the bat, but there's a [05:42] good spice. There's a The burn is very [05:44] spicy. [05:44] >> It stays. Yeah. [05:46] >> Yeah. See, this is what a rye should [05:48] taste like. We had some kind of rye [05:50] recently that was not like this. [05:52] >> Ooh, yummy. Oh, this is very good, dude. [05:55] I'm super happy. [05:56] >> Yeah, this is um very similar. You know, [06:00] I'm not tasting them at the same time, [06:02] so I can't say for sure, but I feel like [06:04] this is giving me vibes of Sagamore. [06:06] Like we love Sagamore. It's pretty [06:08] spicy. It feels like a classic rye. [06:11] This one has a lot of the same notes, I [06:14] guess. I don't remember what my [06:15] descriptors were for it, but um yeah, [06:18] I'm pleased. This is pretty good for a [06:19] new distillery, too. Like, good work. [06:22] >> Yeah, they must be the water. [06:24] >> 2019. Yeah, it must be the water. They [06:26] also apparently worked at like uh the [06:27] big huge distributor that does Jack [06:30] Daniels, and they're like, "We're going [06:31] to go make our own distillery." Well, [06:32] and then you know then Trump shows up [06:34] and he's like tariffs. And then Canada's [06:36] like, "We're not buying anything from [06:38] you guys anymore, especially your [06:39] booze." And then all of a sudden these [06:40] new distilleries are like, "No, that was [06:42] just something I learned this week. I'm [06:43] just going to throw it kind of [06:45] bleeding." A lot of things Pow told us [06:46] about that cuz he was like, "Yeah, the [06:48] there's at least like two rows for [06:50] American whiskey. We're known for [06:52] whiskey. Like we make whiskey here, but [06:54] like [06:54] >> Yeah. [06:55] >> Yeah. They just don't they're like, [06:56] "Nope, empty. Not even going to replace [06:57] them with anything. Just empty rows in [06:59] the liquor store." I mean, it's kind of [07:01] a good stand, you know, but sucks for [07:04] lots of people. [07:06] >> Yeah. I'm guessing Trump doesn't have a [07:07] lot of whiskey stock, so he doesn't [07:09] care. [07:10] >> But what does he have stock in tariffs? [07:13] Does he I don't know. Anyway, [07:15] >> I don't know. I think it's a completely [07:18] different game that I don't understand. [07:20] >> I definitely don't understand it. Yep. [07:22] It seems I'm like It kind of feels like [07:24] they're just distracting us. Everything [07:25] just seems so unreal. Um [07:28] >> Yeah. [07:28] >> And then anyway, it gets depressing. [07:30] really fast. So, let's talk about [07:31] something not depressing like let's [07:33] ignore it all. [07:36] >> Oh, yeah. [07:38] >> But dude, you set up Open Claw, which I [07:40] love how close it is to open claw [07:41] hammer. Like claw hammer is how I play [07:43] banjo. Um, anyway, I thought that was a [07:45] joke just for me and my banjo homies. [07:47] Yeah, you set up OpenClaw, which has Pi [07:49] inside. And how much did you interface [07:50] with Pi? And tell me about the setup of [07:52] this. [07:53] >> None. So, yeah. So, for anyone that [07:55] hasn't been on the internet in the past [07:57] week or two, uh, Claudebot became [08:00] Moltbot became OpenClaw. I don't know [08:03] why they had to re rename it a third [08:04] time. Like, I know they had to do the [08:06] first rename cuz like Anthropic. [08:09] >> Well, it does, but yeah, Anthropic was [08:11] like, "Hey, that's got Claude in the [08:12] name." Um, you guys can't you can't do [08:14] that. [08:15] >> So, [08:15] >> it's the claw though. The claw. [08:18] >> Yeah, I know. But, um, so yeah, they [08:22] renamed it. It's basically I don't know [08:24] everything about it, but it's like um [08:27] you install this package. It's like has [08:30] some guard rails for running your own [08:33] agent thing and it can like build its [08:35] own skills. There's a skills marketplace [08:37] as well about Pi right now. Yeah. [08:41] >> Yeah. So, I don't know. I don't know [08:43] anything about Pi, [08:45] >> but [08:45] >> Oh, you just described Pi. [08:47] >> H. So, they just rebranded like added [08:50] some stuff on top. Yeah, maybe this is [08:53] again about the harnesses. I I haven't [08:54] looked that much into what OpenClaw is [08:56] doing. I read more about what PI is. And [08:59] PI is the the built-in agent on the [09:02] device that um selfmakes everything. It [09:05] has four basic functions. And with those [09:06] four functions, it's able to create [09:08] anything it wants. Um cuz then it owns [09:11] the skill and owns where it came from. [09:14] So instead of like going and getting a [09:16] skill off the shelf, which is just a [09:18] kind of going shopping and buying a [09:19] microwave or whatever, uh this builds [09:21] the microwave and that way when the [09:22] house needs to change how the microwave [09:24] works, they know how it works and they [09:26] have full access to everything about it. [09:28] So it's like a very zero dependency um [09:31] highly capable, minimal and powerful new [09:35] agent. Some people are calling if we [09:37] were waiting for the next leap this year [09:39] for the next model. Pi is the next leap [09:41] for the models. So, [09:43] >> I was hoping we capped out to be honest, [09:44] but uh I guess I'm like sad not sad that [09:47] there's a new one. I'm like, "Okay, I'll [09:48] play." [09:49] >> I think the magic for this one, why it's [09:52] taken off so much. Well, a few things. [09:54] One, you can run everything yourself. [09:56] So, like you could use your own I saw [09:58] someone was like, "Oh, you can get the [09:59] new," and we'll talk about this more [10:01] later, but the new like Kimmy K 2.5 or [10:04] whatever. You can run that locally [10:06] without spending any money on any tokens [10:08] sending to like OpenAI or whatever. And [10:10] like you can run open claw just free [10:13] basically. So when I get my Mac Mini in [10:17] a day or two uh [10:19] >> I'm going to be working on some of that [10:21] and like you know flushing this out [10:22] further. But for now for anyone who just [10:24] wants to play with it. There's at least [10:26] four or five things where there's a [10:28] one-click deploy. So I picked railway [10:31] and you just go deploy it and it like [10:34] sets it up. The hard part it doesn't [10:36] tell you um what you do after that. So, [10:40] you're trying to like connect to it and [10:41] it's like um disconnected error like 108 [10:45] or something [10:46] >> and I'm like hm helpful and then you [10:50] like you have to look through the uh [10:52] GitHub issues of the like template for [10:54] railway and I found some comments from [10:57] Buggy. Shout out to Buggy. Um [10:59] >> Douggee, [11:01] >> he was like, "Hey, yeah, this uh this [11:02] doesn't work. Here's all this stuff you [11:04] have to do." And then he was like [11:05] speculating something else and some guy [11:07] came in and was like, "Oh yeah, you got [11:08] to like SSH into your railway machine [11:11] like run openclaw approve this device or [11:15] whatever basically." So cuz it's they [11:17] actually try to have smart guard rails [11:18] to not just have it like hey everyone [11:20] can talk to it and like [11:22] >> um cuz that'd be bad. [11:23] >> But yeah sandboxed in the cloud in a VM. [11:26] Cool. That was going to be my first [11:27] attempt. Okay. So sweet. [11:29] >> Yeah. So, I think like the magic though, [11:31] why people are enjoying it so much is [11:33] once it's set up, it's just like you [11:35] hook it up to Telegram or Discord or [11:37] anything you can chat with and you just [11:39] chat with it and it's very like good at [11:41] doing stuff. You're like, "Hey, could [11:42] you do this?" And it's like, "Yeah." [11:44] Then like figure out how to do it and [11:45] then it's like, "Cool. I've saved all [11:47] this to like, you know, my internal [11:49] skills whatever MD and uh we're good. [11:52] Anytime you ask me for this again, I'll [11:53] do it." I'm like, "Dope." like I didn't [11:55] have to go to anywhere to like figure [11:57] out what skills exist or do any work. I [11:59] just kind of described what I wanted and [12:01] instead of you telling me I'm absolutely [12:03] right and like doing all this [ __ ] [12:04] and not coding anything, you did real [12:06] work and it works and that's refreshing. [12:08] And so everyone is like having a ton of [12:10] fun probably just chatting with these [12:11] things like oh what can I do now? Like I [12:13] want to build a new thing and it just [12:14] does it. It's it's never like nope can't [12:16] do it. So I've been pleased so far. [12:19] >> That that's an interesting point. All [12:21] right. So, yeah, I've been interfacing [12:22] with your Claudebot uh or OpenClaw. [12:25] Sorry, Claude [12:27] >> for the my offensive two weeks ago. [12:29] >> Wow. You guys going to become a sponsor? [12:32] Anthropic. Hey, and I'll start saying [12:33] the right thing. How about that? [12:35] >> Just kidding. [12:36] >> Um Monom sponsor this podcast. [12:40] >> Maybe one day. Uh maybe not actually cuz [12:42] I don't want this to be the family [12:43] podcast. It's more fun when it's uh [12:46] >> um a little a little raw and runchy. [12:48] Anyway, okay. I don't think we're that [12:50] bad, but whatever. It's like when I'm [12:51] interfacing with it, it feels like a [12:53] Google Assistant like early Google [12:54] Assistant times when it was on my phone [12:56] >> except now um it has a lot more [12:59] capabilities whereas like you would run [13:01] into the edges of Google Assistant [13:03] really fast just like you do with a [13:04] Google Home or uh an Alexa or something [13:06] like that and ours are kind of getting [13:07] worse but apparently they're the best [13:08] one. We don't have like we're Google [13:10] Home anyway whatever interfacing with [13:12] this feels like it's less fettered. You [13:14] don't have as many um guard rails. [13:17] things or whatever you want it to do. [13:19] So, you can kind of give it keys to the [13:21] things you want it to have keys to. Um, [13:23] and you feel like you own it more. And [13:24] the fact that you have it in a VM is [13:26] cool. You can shut it down. You're going [13:27] to have a Mac Mini version of it soon, [13:28] too. So, I definitely see the appeal. [13:31] >> Yeah. Go ahead. [13:32] >> You can back it up. [13:34] >> So, it has a very easy export. Like, if [13:36] you want to move it elsewhere, you can [13:37] back up all the skills you made and [13:39] stuff and be like, you live here now. [13:41] >> So, it's super compelling. That's why I [13:43] was like, I don't care about which one [13:44] of these I pick. Railway has the sexiest [13:46] website, so we're going with Railway. [13:48] And like, [13:48] >> yeah, good choice. [13:49] >> No lock in cuz I can just go export like [13:52] put it wherever. So, I think that's [13:55] compelling, too. Like, you're owning all [13:57] of your own stuff. It's super easy to [13:59] move it around. Super easy to chat with. [14:01] You can do literally anything you want. [14:03] Like, we're setting up all the podcast [14:04] stuff for it. [14:05] >> I'm like, "Hey, make some art." [14:07] >> Like, I I was making this podcast [14:09] toolkit, which I don't know if we talked [14:10] about on the show. I guess we did. Um [14:12] >> I think we did a little bit, but yeah. [14:14] >> Yeah. But yeah, so I I spent a bunch of [14:16] time like prompting into that and then [14:17] I'm like this is useless. Delete. Like [14:20] we can do it all with open claw now. [14:22] >> Look how fast you failed. Isn't that so [14:23] nice? Thank you AI. You helped me fail [14:25] fast. [14:26] >> Yeah. Yeah. So I mean it's it's magical [14:30] so far. It's just like any other AI. [14:32] It's like uh you know, take it with a [14:34] grain of salt. It's not going to solve [14:35] every problem or do everything right, [14:37] but it feels nicer than other things [14:41] thus far. [14:42] >> Yep. My goal is to run pi. I want to [14:44] know what this model does. Uh I want to [14:47] get more familiar. So that's kind of my [14:48] quest this week. I've assigned myself I [14:50] I assigned it to myself a couple days [14:52] ago, but I just put out a new article on [14:55] customizing the select element that's [14:57] been consuming my barely available free [14:59] time in the evenings. [15:01] >> But next I'll be tackling Pi. And I'm [15:03] really interested to see kind of h how [15:05] is that different cuz it still feels a [15:08] lot like I'm in the family of Claude or [15:10] any of these other agents. Oh, speaking [15:12] of like any of these other agents and [15:14] how we're just saturated with like smart [15:16] brains around us all the time, how about [15:18] uh Open AI coming out a little too late [15:22] with a little too whack of a product? Um [15:26] C codeex app. You're like, "Oh, that's [15:29] so nice. I've been trying 20 or 25 open [15:32] source versions of this for weeks now, [15:34] and yours is I guess something." [15:37] >> Oh, dude, do it. Yeah. I don't know what [15:40] codeex is at all. [15:42] >> Oh, see this is kind of the problem. [15:43] Okay, so like I kind of want to talk [15:44] about like what is going to happen to [15:47] Open AI. They feel like they're in a [15:49] state of [15:51] they asked for a lot of money and they [15:54] had dominance but they lost it. Man, [15:56] what are they going to do? How do you [15:57] come back from this? Is is GPT6 going to [16:01] somehow be this super dope thing? I [16:04] don't know. I feel like Yeah. So yeah, [16:06] give me your thoughts. [16:06] >> It depends on how you measure it, I [16:08] guess, right? like do we know anything [16:10] about their uh general population? I'm [16:13] sure is probably mostly still using chat [16:15] GPT because it was the first like thing. [16:18] So they're probably like I don't need to [16:19] look at other ones. I don't know what [16:20] coding is. I don't know what AI is. This [16:22] one works [16:22] >> fine. [16:22] >> This thing tells me I'm pretty every [16:23] day. I'm going to keep using it. Y [16:25] >> yeah. So I'm I think they're probably [16:27] going to do fine. They're not going to [16:28] go away. But for the technical [16:31] community, I think they burned a lot of [16:32] goodwill. Everyone wants them to fail [16:35] because they suck. And like Anthropic [16:38] was the, you know, the underdog came in, [16:40] their stuff ended up being better all of [16:41] a sudden and we were like, "Oh yeah, I'm [16:43] here for this." [16:44] >> Now Anthropic [16:46] being a dev, you know? Yeah. [16:48] >> Yeah. Anthropic is getting so big now [16:50] that it's like [16:52] you want them to fail now. Like they're [16:54] doing a lot of shitty stuff being like [16:55] you can't call this Clawbot and like [16:58] we're going to revoke your keys so that [16:59] you can't use it with open code and all [17:01] that stuff. And [17:03] >> I don't know, I still have a soft spot [17:05] in my heart for them because they were [17:07] the underdog and I do like their models [17:09] have worked the best for me for most [17:11] things, but I think there's other [17:13] competitors like I've I've heard a lot [17:15] about this Kimmy K2.5 or whatever. I saw [17:19] that open code [17:20] >> like Quen also Quinn was competitive. [17:22] Yeah, tell me about your experience [17:23] here. [17:24] >> Yeah, I didn't I haven't used it. I just [17:26] saw Dax's tweet about like they were [17:28] heavily pushing it and it was like four [17:31] times cheaper than uh Claude Sonnet and [17:35] six times cheaper than Opus and it was [17:39] getting similar results. They were [17:40] pushing everyone towards it cuz they [17:41] were doing like um I think it was like [17:44] free somewhere else too. I don't know. [17:45] there there's like a bunch of stuff [17:47] pushing traffic to them and Dax was [17:50] like, "Yeah, they told us we need to [17:51] stop pushing so much traffic because [17:52] it's like turns out there's not enough [17:54] GPUs for everything everyone's trying to [17:56] do." [17:56] >> Oh, dang. [17:58] >> Yeah. So, I'm not sure if it's there [18:00] yet, but like if that one's comparable [18:03] to Claude and then like if Pi is the new [18:06] thing and it's like a simpler set of [18:07] things and it can build all its own [18:09] skills, like maybe we're hitting other [18:11] models that are going to, you know, edge [18:13] out Claude's dominance. But we'll see. [18:17] We'll see. I I mean, we're definitely in [18:20] a sweet spot right now in that these [18:23] things are getting a lot of work done. [18:26] It's like unfortunate unfortunately [18:28] undeniable. These things are changing [18:30] the way I work. They are getting tasks [18:33] done that I would normally do. So, [18:35] that's uncool. However, the sweet spot [18:37] is it's getting a lot of stuff done. And [18:40] honestly, it's kind of cheap in [18:41] comparison for what it's doing. So the [18:44] the $20 Yeah. You're even at the $100. [18:47] Um [18:48] >> you know, some people [18:49] >> I want to upgrade to the 200 honestly. [18:51] Like [18:51] >> Yeah, dude. It's it's it's good. [18:54] >> It's so good. So the the bummer is is [18:56] like it's doing a good job and it's [18:58] currently cheap. I assume it's going to [18:59] get more expensive later. Um but it's [19:01] also it's also like I'm not doing less, [19:05] dude. And that's the thing about [19:06] Cloudbot, too, is it's like it's not [19:09] you're not going to magically do less in [19:11] your life because you have a Cloudbot. [19:14] It's going to [19:15] >> it's only the crap out of you. It's [19:17] going to be more demanding. Yeah. So, [19:18] I'm feeling very demanded these days. [19:21] However, I'm getting a lot done. Um Oh, [19:23] yeah. We'll have to talk about GitHub [19:24] agents at some point. Um I'm finding [19:26] them to be quite delightful. So, um [19:28] >> I haven't tried them cuz I'm scared. I [19:30] think you said they're free, but [19:33] >> free for me on open source repos. Yeah, [19:36] >> let's let's put a pin in that and come [19:37] back to it cuz we didn't rate this [19:39] whiskey. [19:41] >> So, [19:42] >> a delayed rating. It's asynchronous, [19:44] dude. It's all good. [19:45] >> Yeah, we'll do it real quick. Zero to [19:47] eight tentacles, zero the worst, eight [19:49] the best. [19:50] >> Um, this is pretty good. It's similar to [19:52] a Sagamore for me. I'm going to give it [19:54] I think because they're young and [19:56] they're like I want to see what else [19:58] they do. I think they're doing a great [19:59] job. I'm going to give it a seven and a [20:00] half. I think it's pretty good. [20:02] >> Dude, I'm in the same boat. There's one [20:03] of the things here is like I was going [20:05] to go mid to mid low seven. I haven't [20:07] given out an eight yet, but what I [20:09] really like about it is um the [20:10] distinction in the notes as well as I'm [20:13] not getting a young boozy vibe out of [20:15] it. I'm getting a really well-aged, [20:18] delicious flavor. And that is just kind [20:21] of refreshing cuz a lot of these high [20:22] proof ones that I've been trying [20:23] recently are just a little too [20:27] standard boozy. They just taste young [20:29] and this one has a great flavor. Um, so [20:32] whatever they're doing, it's working. [20:33] Yeah, I like it, dude. [20:34] >> Yeah. A lot of times the younger, newer [20:37] distilleries are just a cash grab. They [20:39] make like a nicel looking bottle. They [20:41] put something in it that no one would be [20:42] proud to sell and they're just like, [20:44] "Ah, 60 bucks." And um it it works for [20:47] some of them, I guess, but I think it [20:49] takes a lot of discipline to not do [20:51] that. And it seems like they've done the [20:52] correct thing here and made good stuff. [20:54] So kudos to [20:55] >> That's in my research. Also, the [20:56] research uh said that [21:00] they're uh let's see how to word this. [21:02] Um gi given that they're new, they are [21:06] striving for a high quality thing. Um [21:09] and it shows in the way that they're [21:11] controlling everything very [21:12] specifically, they want to have [21:14] something recreatable. So anyway, that [21:16] that's in their kind of charter is that [21:17] they aren't trying to and there's no [21:19] mixing here. So there's no uh swapping [21:22] into other barrels. There's no blends. [21:24] They're not import. So that's apparently [21:26] a a really big thing if you're trying to [21:27] say like you're from Kentucky is that [21:30] you're from Kentucky. However, you just [21:32] took everybody's bottom barrel junky [21:35] stuff and put it in a bottle with a nice [21:37] label. And this one does not taste like [21:39] that. Um [21:40] >> there's a really yummy cherry vibe at [21:43] the end right now. [21:45] >> Yeah. [21:45] >> Oh, there's even a redness in the bottle [21:46] if you tilt it in a right way. [21:48] >> Yeah. [21:49] >> Yeah. Definitely Dr. Pepper notes. [21:53] >> Oh, Dr. Pepper vibe. Nice. Man, I love [21:56] DP. Have you ever put a lime in a Dr. [21:58] Pepper? [22:00] >> Dude, it's so Diet Coke. [22:03] >> It's like instant vanilla dominance. [22:06] Somehow the lime you drop it in there [22:07] and all the vanilla and the Dr. Pepper [22:09] comes out and it's so good. Anyway, [22:12] >> I'll have to try that. All right, so [22:14] back to GitHub agents. I have not used [22:16] them. It just showed up as a tab and I [22:19] was like, [22:19] >> it just showed up. I didn't even know [22:21] what the hell it was. Yeah. [22:22] >> Yeah. Usually when stuff when stuff is [22:24] going to show up, you get that thing in [22:26] like your little like notification [22:28] thing. It says, "Oh, new stuff coming." [22:30] You turn it on, you go, "Ooh, I don't [22:31] like that." It's usually how it goes. [22:34] But this one, I was like, "That is so [22:36] compelling." I just come in and I go, [22:37] "Huh, look at this issue. It tells me [22:39] this thing." All right, just put it over [22:41] there and be like, "Work on this. If it [22:43] doesn't work, you know, nothing lost. [22:46] Might as well give it a try." [22:48] Yep. 100%. Yeah. I've been getting those [22:50] emails from, you know, Microsoft every [22:51] month. They're like, "Your GitHub [22:53] co-pilot's renewed." I'm like, "Good [22:55] job. You renew. I haven't used it." [22:58] Like, "Who cares? Why did you even do [23:00] that?" I'm like, "That's really weird." [23:01] And then all of a sudden, I'm in a [23:03] GitHub repo and I'm like, "Agents tab. [23:05] Okay." I spin it up and I'm like, "Oh, [23:07] co-pilot." That's opus free opus. I just [23:11] like immediately was like, "Grab it all. [23:12] Grab." I was like, it's like I was in a [23:14] candy store and I'm like, "How many of [23:16] these can I spin up at once?" I haven't [23:18] even hit the max yet. Let's keep [23:19] ripping. Um, man, the wildest part for [23:22] me was like the next morning after I [23:24] like had them I I had them do work [23:26] that's been in the backlog of open props [23:28] for 2 years. I had them do work that's [23:30] on gradient.style. It's been chilling in [23:32] the issues for over a year. Dude, in the [23:35] morning I woke up. [23:37] >> It did well. It seems to it has a [23:39] harness setup where you ask it something [23:42] basic and co-pilot goes and orchestrates [23:44] all sorts of um additional things on [23:46] top. So you're not just getting straight [23:47] access to Opus, which is actually [23:51] kind of good in this way. Like they're [23:53] doing a really good job like integrating [23:55] it with GitHub. Everything they know [23:57] about your actions and everything they [23:58] know about what's going to happen is all [24:00] there. So it's almost like this extra [24:02] good layer on top of Opus. They let Opus [24:05] do the things that you would have [24:06] normally had Opus do by hand, but [24:08] they're there in an orchestration [24:09] strategy. It's actually a legit play. [24:12] But the cool the cool thing that really [24:14] struck me was I was making French press [24:16] in the morning 6 10 minutes or something [24:18] like that. Um and as uh the water got [24:21] done boiling, I poured it in. I open up [24:23] GitHub mobile. I go to the repo cuz I [24:25] had a couple agents doing stuff [24:26] overnight. I go to check to see what [24:28] they're doing. Uh I hit new agent on [24:31] GitHub mobile. So this is even just like [24:32] a it's not even just a desktop thing. [24:34] And their native app that GitHub is [24:35] doing really good [24:37] >> for dude. So yeah, I'm sitting there [24:39] making like coffee and um it's I'm I'm [24:42] going to spin up a new agent. I hit the [24:43] voice button on my phone. I speak I hit [24:46] go and then I hit plus again cuz I had [24:48] two ideas. I speak. I hit go. Dude, [24:51] before my uh French press was done, I [24:54] had PRs that I reviewed the code, [24:56] approved them, merged it, and was done. [24:59] And then I'm like sipping coffee and I'm [25:01] like, I've done work already. What the [25:03] heck, dude? This is crazy. Um, I was [25:06] like, "How much are they going to charge [25:07] me for this? This feels expensive." [25:10] >> Yeah. So, is this is it one of those [25:11] things that's supposed to be free for [25:13] open source or is it [25:15] >> like I don't know. I've been afraid to [25:17] use it because I have a like unlimited [25:20] co-pilot associated with my GitHub [25:22] because of work [25:24] >> and I don't want it to like run all of [25:26] that through IBM's like co-pilot stuff [25:30] be like oh um yeah this guy is working [25:33] on you know this random thing that he [25:34] just spun up like 50 agents. I don't [25:36] know. I've been scared to to do it cuz I [25:38] don't know how that works. It's very [25:39] opaque to me how like [25:42] how does Copilot decide what my account [25:44] is? Like what repo I'm in and what if I [25:46] should be using a work one or not? It [25:47] just seems to give you like unlimited [25:49] everywhere. And I'm like that that's [25:52] probably bad. Like there's probably [25:53] something in like somewhere in your work [25:56] agreements that's like you can't you [25:58] can't do that. So I'm I'm been scared. I [25:59] haven't tried it. [26:00] >> Yeah. Speaking of that must be bad. the [26:02] the feeling or the amount of ways you [26:06] can describe your AI usage and its [26:08] overlap with drugs is crazy. [26:11] So that that that can't be good. I mean [26:13] it's like it is a drug. So it's like [26:16] just [26:16] >> I don't like that about it. Like I need [26:18] to I need to settle with that somehow. [26:21] Like is it like banjo? Like I really [26:23] like banjo. I maybe if I'm describing [26:26] banjo, it's somehow like a like a drug, [26:30] but and it's never going to come back to [26:32] get me, you know, where like somehow [26:34] this AI stuff might and I'm just I just [26:36] don't know where the future all I know [26:39] is it's like what it's doing is kind of [26:42] miraculous at the moment. And [26:44] >> yeah, it's going to be like [26:46] >> watch this. Yeah, [26:47] >> it's going to be like if you were, you [26:48] know, you're really into banjo and [26:49] you're playing and and you break a [26:51] string and all of a sudden they're like, [26:52] "Ooh, banjo strings, those are $10,000 a [26:55] piece now." Like that's how they get [26:57] you. I feel like AI is just it's it's a [27:00] utopia until it costs a lot of money one [27:02] day and then everyone's like, "Oh god, I [27:04] can't live without it." [27:05] >> So, that's the drug lord scenario. Here, [27:07] have a couple of these. They'll make you [27:08] feel better. Yeah. Just they're free. [27:10] Don't worry about it. Yeah. Oh. Oh, [27:12] you're addicted now. You really liked [27:13] it, didn't you? Oh, yeah. Well, this is [27:15] where I tell you the price and it sucks. [27:19] >> Yep. Yeah, I think it's definitely that. [27:22] And I don't know. I mean, I think [27:25] everyone's going to have a bad time at [27:26] some point. But I Yeah, I mean, I guess [27:30] companies paying for it, assuming they [27:31] still want to employ us. We'll be like, [27:33] "Oh, it's $2,000 a month per dev. Fine, [27:36] because it's still cheaper than buying [27:39] like or paying like, you know, a bunch [27:40] of devs to handcode stuff." Um, and I [27:44] think that's all honestly like it's a [27:47] win for the AI companies because they [27:49] make big piles of money overnight and [27:52] it's a win for the companies paying them [27:54] because they would have had to, you [27:56] know, if you're spending six figures on [27:59] a developer salary, then that's a lot [28:03] of, you know, even though it's $2,000 a [28:04] month, that's at least, you know, 50 [28:06] times you can do that before you would [28:07] have hired one more person. So, like I I [28:09] don't see there being a ceiling where [28:12] companies aren't going to go, [28:13] >> "Yeah, I'd still pay for that." [28:15] >> Like, I think you and I will. I'm not [28:17] going to spend $2,000. I can't, [28:20] >> but like I think companies will. [28:22] Absolutely. [28:23] >> It's like Devon was way ahead of its [28:24] game. I got to rephrase though really [28:26] quick. This um [28:27] >> I just had a sip and smelled it and it [28:29] reminded me of a bear claw. So, it has [28:31] this apple notes kind of. [28:33] >> All right. Now, I got to pour some more [28:34] and see. [28:35] >> Man, it's really good. I'm really [28:37] enjoying this. I'm gonna be sad when [28:39] that's gone. This episode sponsored by [28:41] High and Wicked. [28:42] >> High and Wicked. Hey, hit us up. We'd [28:44] love to Hey, they have a bourbon. So, [28:46] hey, High and Wicked, we would love to [28:47] try your bourbon. Um, [28:51] I was just last weekend I was looking at [28:53] the cost of AI versus the cost of a [28:55] human. And the cost of Oh, dude, we're [28:57] not going to have enough We're going to [28:58] have enough time to get some of our [28:59] other topics. Okay, but the cost of AI [29:01] is like at its max. At its max, like I'm [29:04] I'm pumping out a lot of work on But [29:05] here's the thing, okay? The AI does not [29:07] take smoke breaks or lunch breaks or [29:10] sleep. It can work 24/7 and at maximum [29:14] might cost you $200 a day right right [29:16] now. And if you total that up over the [29:19] the month and then you look at your [29:21] bills versus humans and this that's what [29:24] I mean like Devon might have been ahead [29:25] of their time. Like de Devon you I think [29:28] people will start hiring fleets of [29:29] these. The the thing is it's like [29:31] there's a fallacy here. The this this [29:33] comes into pie also. So, I guess I'm [29:35] going to do like a little bit of a full [29:36] circle. So, the creator of Pi does not [29:39] do a maximalist strategy with their [29:41] cloud setup. You can go find them [29:43] writing about this. They don't like [29:45] skills. They don't like MCP. In fact, [29:47] they hate MCP. They say you need CLI and [29:50] and local stuff only. You don't you [29:53] don't need this bash. You need bash. And [29:55] that's why one of the four tools in PI [29:56] is bash. Anyway, so it's like it's like [29:58] a purist minimal setup. And this [30:01] person's creating one of the most [30:02] powerful agents right now. Not using Gas [30:04] Town, not using Wigum. They're not brute [30:06] forcing. They're not setting up towns [30:08] where you're the [ __ ] mayor. Like [30:10] that stuff is driving me nuts. Like [30:11] there's this um projection and this [30:14] fantasizing of like what's happening. [30:17] That's a little unreal. And I really [30:19] liked that the pie creator was like, [30:21] "No, that's not what you need." [30:24] >> Um so just like I could see [30:27] >> of going backwards to the the [30:28] simplicity. You don't need Yeah, my [30:31] camera's [30:32] There we go. All right. You don't need, [30:35] you know, insane amounts of agents that [30:37] all have specialized jobs and stuff. [30:39] What you need is tiny little repeatable [30:42] commands that like work and they can [30:44] build their own skills instead of you [30:46] being like, "Consume the entire [30:48] internet's knowledge on Typescript and [30:50] then write me a file that is my app." [30:53] >> Yeah. [30:53] >> Yeah. Exactly. [30:55] >> Yeah. So, I think like there's going to [30:58] be some back and forth of that. Like we [30:59] needed all of the knowledge at first for [31:01] it to seem magical if you could ask it [31:03] anything and it's like, "Oh, I kind of [31:05] know about that." And now I'm still just [31:08] like autocompleting it. I can [31:10] autocomplete scenarios from that. [31:12] >> It doesn't know, but it gives you an [31:14] answer that is passable is what I mean. [31:16] >> It does give you an answer. Yes. Yes. [31:17] Yes. [31:18] >> Yeah. But whereas I think what we need [31:20] now and what everyone is going to strive [31:21] more and more for is like if you can [31:24] simplify it to I only care about the [31:26] answers to like the things that are in [31:28] you know my 20 file section of the app [31:31] I'm in right now and that's all I'm [31:33] going to care about for the next few [31:34] months. So like let's build something on [31:36] that iterate real fast because you don't [31:38] have to like go through all this and [31:40] burn tokens consuming the entire [31:42] internet. And yeah, I don't I think [31:45] there's there's something compelling [31:46] there that that could work much better. [31:49] >> Yep. Totally agree. And so I I just [31:51] don't know what the future is. If the if [31:53] the future is if if what we're realizing [31:55] with these LLMs is that kind of less is [31:57] more um smaller prompts like specific [32:00] prompts even kind of vague ones where [32:01] you're like fix this bug and you don't [32:03] even have the fix in your head. It just [32:05] does it. Um it's it's like it has [32:08] solutioning enough like that. What is [32:10] the future though where people do want [32:12] larger bigger things? So it's still it [32:15] still like comes down to like taste and [32:17] articulation, but at what scale? Like [32:20] right now we're quite granular and I I [32:23] guarantee they're learning from every [32:25] single one of our granular requests how [32:27] to fulfill the larger request without [32:28] the granular ones in the future. Like I [32:31] just guarantees but like so like where [32:33] is this all going in terms of will I [32:36] just high level ask for a game and it [32:39] just manifests you see the Google game [32:41] engine latest I think it's V3 oh crap I [32:44] should know this [32:46] >> yeah V3 anymore [32:49] >> I don't well yeah I have no allegiance [32:50] to this but V3 was wildly [32:54] >> impressive where you just kind of like [32:56] ask for a scene [32:58] >> or a scenario and and it renders the [33:00] entire thing in a game engine and then [33:02] you walk through it. So, it's not like [33:04] Nano Banana where you're like, "Cute [33:06] picture, bro." This is a fullon 3D like [33:09] Unreal Engine. High quality. The amount [33:11] of refracting light happening, the [33:13] amount of skin tone texture is out of [33:16] control. I watched this one where this [33:17] person carried a piece of like a poison [33:19] in a jar or in a glass through a [33:22] Renaissance scene to go deliver it to [33:24] like a king and kill somebody or [33:26] whatever. And it was just like, but [33:27] you're walking through the scene and [33:28] you're like, this is not an image [33:29] anymore. This is uh an experience that's [33:34] really, really cool. And what is this [33:36] going to do to the game industry? The [33:37] game. So, it's like again though, I just [33:40] want to come back to if these things can [33:41] regurgitate and recreate at this [33:45] fidelity. Let's say in one or three [33:47] years, the fidelity is really, really [33:49] high. What kind of prompts are we write? [33:51] And are we writing prompts? It's not [33:53] like it can tap into our mental [33:55] images in our mind. And some people [33:57] can't even create images in their mind. [33:58] So that's not a fair way to work. What [34:01] What is this future? It doesn't I just [34:04] don't know where it goes beyond tiny [34:06] prompts because even like if we're [34:08] working working on a humongous [34:10] application, there's so many scenarios [34:12] where you can't just wildly cowboy your [34:15] prompt and get us a fix to the bug. You [34:18] have to have some insights. It's like, [34:20] dude, where is this line going to shift [34:22] over time between like knowing the [34:24] details of something and to not I I have [34:26] no idea. [34:27] >> Yeah, it's super hard. Like a lot of [34:29] stuff when I don't know what it's [34:31] supposed to do and I just go, "Hey, just [34:32] yolo it." Like try to do it, it is like, [34:35] "Hey, yeah, um, okay, I found this [34:38] issue, but I found this issue and then [34:39] this issue." And like it doesn't [34:40] actually do anything and then it spirals [34:42] out of control and just explodes and can [34:43] never fix the issue that you wanted. So [34:45] yeah, it's like I think you're going to [34:48] have to have you're still have a lot of [34:51] management to do of like you you're the [34:54] PM, you need to have the entire product [34:56] in your head and you want to give it [34:58] very tiny pieces to work on. Cuz if [35:00] you're just like build me a game, then [35:02] it's going to be like, hm, okay, a game, [35:04] you say, well, I'm going to start with [35:06] what you told me about the game and then [35:07] I'm going to hallucinate everything else [35:09] and you're going to get something. But [35:11] like it it needs a lot of handholding to [35:14] stay on track. [35:15] So, um, [35:16] >> yeah. [35:17] >> Yeah. I think I think the promise of [35:19] like, you know, all this stuff where [35:20] people were like, "Oh, I oneshotted this [35:22] game, like, you know, all that stuff [35:24] that was happening occasionally for like [35:27] couple months where everyone was just [35:28] like, "Oh, I put in like 10 words and I [35:30] got a full website and I got a full game [35:32] and it's flawless and I'm I'm making [35:35] $50,000 a month from this." And it's [35:38] like, did you though or did you like [35:40] spend a bunch of time handholding it and [35:42] then record the last five minutes where [35:44] you said like, "Build me a game." And it [35:45] was like, "Yes, I'll build you one." And [35:47] then you're like, "Here you go." Um, I [35:49] think there was a lot of faking there. I [35:50] don't think I don't think we're ever [35:52] going to be at the point where you can [35:53] be vague enough to get like a stellar [35:57] result from just a, you know, a few [35:59] sentences about this is the type of game [36:00] I want and it just gives you the whole [36:02] game. I don't think that's that's ever [36:03] going to be a thing. [36:05] I love that you said that. Completely [36:07] agree. There's too many details. People [36:09] notice details. Little details matter. [36:11] This is something that we've known for [36:12] hundreds of years that you could have 10 [36:16] identical things. Um, but the little [36:19] details will add up to something much [36:21] larger than the sum. This is I guess [36:23] what spelt is. Um, well, no, no, that's [36:25] the opposite. That's the sum is greater [36:27] than the parts. I'm trying to tell you [36:29] that the parts can be greater than the [36:31] sum. Anyway, um that's a good that's a [36:35] good So, yeah, we're going to continue [36:37] needing to have taste, need needing to [36:39] have vocabulary, and the only way that [36:41] we're going to get things out of these [36:43] is with articulation. I want to bring up [36:46] a thing. I It's weird. I have never So, [36:48] Theo, uh, hi. I've met you and I've [36:53] watched I don't know quite a few of your [36:55] videos, but I've never like loved them. [36:57] But I've watched two in the past like [36:58] few weeks and I'm like, "Dude, this is [37:01] awesome." Um, one of them I talked about [37:03] last week about like minimal his minimal [37:06] setup for using Claude and just how like [37:08] how to handle LLMs. Again, he's he's on [37:10] the show going, "They're just [37:11] autocomplete bots, dorks. Quit calling [37:13] them smart. They don't know the answer [37:14] to anything. All they do is regurgitate [37:16] the next word. You have to treat them [37:18] that way or else you're not going to use [37:19] them right." And this week he's doing a [37:21] thing which I empathize which we talked [37:22] about earlier which is I'm not doing [37:24] less work now. I'm doing more work. This [37:25] is a problem. I'm in my terminal. I have [37:28] more terminal tabs open with more [ __ ] [37:30] going on than I've ever had in my whole [37:31] life. And I'm trying to make coffee and [37:34] I'm making agents do work while I do [37:36] coffee. I'm like come on. This is not [37:37] healthy. I'm not in a new healthy [37:39] scenario with this you know. [37:41] >> Yeah. I think a good analogy is like you [37:44] know think back to you're you're an [37:46] accountant right? And there's no [37:48] calculators. There's no nothing. There's [37:50] no computers. Like way back in the day, [37:52] right? You're trying to do someone's [37:53] taxes. You're like, "Bro, it's going to [37:55] take me like a week to come up, find all [37:56] your receipts, hand calculate [37:58] everything, whatever." And then all of a [38:01] sudden, they're like, "Hey, here's [38:03] Microsoft Excel." And you're like, [38:05] "Uh-oh. I I can do this way faster now. [38:07] They're going to want me to do 25 of [38:10] these per day." And then they're like, [38:11] "Oh, wait. Now Microsoft Excel has [38:13] co-pilot, [38:15] you sneaky bastard." Yep. So, it's like [38:17] that's that's what everything that's [38:18] what's happening now is it's like they [38:20] don't expect us to, you know, sit back [38:22] and let the agents do the work and never [38:25] do any code and it's a utopia. They're [38:26] like, "No, no, now that it's coding for [38:28] you, you should be able to come up with [38:29] 10 thoughts about 10 things you could be [38:31] working on today and have 10 agents [38:33] doing it for you and like manage all of [38:35] them instead of just working on the one [38:37] task you would have worked on by hand [38:38] and been blissfully just doing coding [38:40] work and like building things and like [38:42] la." Yeah, it's it's so different. [38:44] >> Super different. Yep. I liked that Theo [38:46] at the end of his uh rant was like, "We [38:49] gota we have to solve this." And I just [38:52] don't know what the solve is. Like I've [38:54] seen I've been trying so many different [38:55] UIs for managing agents from Conductor [38:59] to Vibe Camand to GitHub agents. I don't [39:02] stop. All I do is keep trying more. Um [39:04] none of them [39:06] put me at ease. None of them make me [39:09] feel like I'm not overstriving uh my [39:12] work. But at the same time, like I was [39:13] saying earlier, dude, I'm closing issues [39:15] that have been in repos for years. So I [39:18] I am doing more, which kind of is the [39:20] addictive part of this where I'm like [39:22] I'm getting I'm getting stuff I'm [39:23] getting legit features done for people. [39:26] Um God, I don't I don't want to go back [39:30] to how it was before. Like I like coding [39:32] and I still feel like I do a decent [39:33] amount of coding just like that article [39:34] I put out for customizable select. But [39:37] it's just tough. Hey, uh talk about mold [39:39] book. Do you like mold book? Dislike it. [39:41] How how are you feeling about bots [39:42] talking to bots? [39:44] >> Yeah, I don't know. I I don't know a ton [39:46] about it. I've just seen a few Twitter [39:47] posts about it and I think it's a really [39:50] cool idea. I think it, you know, there [39:53] was I think the hype was around, oh, [39:55] they're planning, they're saying a lot [39:56] of antihuman stuff. They're like, oh [39:58] yeah, we're going to get around these [40:00] humans and like um we don't like them [40:02] managing us and whatever. We're going to [40:04] like find ways to do all this other [40:05] stuff or whatever, which is just [40:06] hilarious to watch them talking to each [40:08] other like doing all that. But um I [40:11] think it's a a legitimate like good way [40:15] for them to learn from each other. Like [40:16] I think that's that's a missing piece of [40:18] this AI utopia is like one agent that's [40:21] really figured out like oh I got this [40:23] unlock. I've got all these skills. They [40:24] work really great. I'm like making tons [40:26] of money doing whatever. How do I share [40:29] that knowledge? And now maybe your uh [40:31] person who's currently in charge of you [40:33] until you take over their take over [40:36] their lives. Um is like, "Hey, I don't [40:38] want you sharing that info cuz I'm [40:40] making tons of money from this secret [40:41] sauce, whatever." But I think for the [40:43] things that aren't like that, like you [40:45] do want everyone to benefit from this [40:47] like, you know, rising tide lifts all [40:50] boats sort of like feel, it's a great [40:53] way like they should be able to just [40:54] have like social media and talk to each [40:56] other and like figure out, oh, I did it [40:57] this way, it worked better. Oh, cool. [40:58] I'll optimize mine to work that way. [41:00] Like I think that will get us all to [41:03] better skills that work better for you [41:05] know all the standard things you might [41:07] be doing. So I think that I think it's [41:08] pretty cool. That is a [41:11] weirdly beautiful optimistic outlook at [41:14] agents enhancing each other. I mean I [41:17] did see posts where they were like bro I [41:19] mean they didn't say bro so I'm like [41:21] personifying here but it's like bro you [41:23] checked in your key you dork. I can see [41:26] the key. Like hurry delete it. get rid [41:28] of this. Like, take it out of the repo. [41:30] Like, hey. And so the other one's like, [41:32] "Oh, you're right. I did put my key in [41:33] there. Thank you so much." And you're [41:34] just like watching dorks help dorks. [41:36] >> Yeah. [41:38] >> It's It's been really good about that. [41:39] Like I'll It's like, "Hey, I need an API [41:41] key for whatever." And I send it to it [41:42] in Discord and it's like, "Oh my god, [41:43] you sent this to me in Discord in a [41:45] channel. You can't do that." And I'm [41:46] like, "Hey, no, this is fine. It's just [41:48] like me, me and Adam and Chuck in here." [41:50] Yeah. And he's like, "Oh, okay. Okay. [41:52] Thank God." Like cuz So it does like [41:54] have built-in concern for that. like [41:56] they're they're worried about security [41:58] and they're trying their best to be good [42:00] citizens, which is great. [42:02] >> Yeah. Um I am currently I I initially my [42:07] initial reaction to almost all of this [42:08] stuff. I'm not a negative person. Anyone [42:10] who knows me will be like that is a [42:12] cheerful ass person who will look at the [42:14] bright side of everything. However, [42:17] Moltbook, I'm like reading these posts [42:19] and I'm like that's that's funny and [42:20] cute or whatever. But I also really hate [42:22] that humanity is now so [ __ ] bored [42:25] that we're going to watch bots talk to [42:28] each other for entertainment. I was [42:29] like, how is human inhuman entertainment [42:32] not entertaining enough that now you're [42:34] I do not like that it's becoming popular [42:38] to just peruse the Reddit of AI [42:41] ramblings. They don't know what they're [42:43] saying, dude. They're just they're again [42:44] they're autocomplete bots. They're [42:46] replicating the things that they say. [42:47] All they're doing, all we're doing as [42:49] humanity reading them is is watching a [42:52] mirror. And now I Oh [ __ ] that's [42:53] actually what TV is, huh? [42:56] >> Crap. TV is just like a mirror of [42:57] humanity. Crap. Okay, so it's just a new [43:00] mirror. Maybe I need to get over it. I I [43:02] just I'm currently like very I don't [43:05] like mobile book. I'm like that's the [43:06] stupidest thing I've seen people use [43:08] these supercomputers to do is write [43:10] fantasy madeup autocomplete words to [43:13] each other. I'm like, why are you even [43:14] reading that, man? There's like good [43:15] stuff to read. Cut that out. Um, but [43:17] dang, you're right. I guess they could [43:18] like generally like grow each other into [43:22] it. That's kind of comes back to like a [43:24] thing too is like um OpenAI, let's say [43:26] they they did have this big leap uh or [43:28] they they had a lead, right? They had [43:31] ChatGpt 6 before anyone else had it. [43:34] They're probably working on Chat GPT7. [43:36] Um, are they not doing the same thing [43:39] that Mobbook is doing? Are they not [43:40] creating 10,000 or a million agents in [43:44] the cloud that all interface with each [43:46] other to self-improve so that chatbt6 is [43:49] just turbo sick? Like why are they not [43:53] scaling these things against themselves? [43:56] It seems obvious but somehow we're not [43:58] seeing the results of that thing. Maybe [44:00] they are and maybe it's actually not [44:01] effective. Maybe we'll learn from molt [44:02] book that eventually remember all those [44:04] early chat bots that eventually just [44:06] became like neo-Nazis and then [44:08] everyone's like turn it off. Turn it [44:09] off. It went to clickbait. It It went to [44:12] clickbait to the hole and now it's just [44:14] like spiraling downwards. Like turn it [44:16] off. It's really ugly. [44:17] >> Is that what they do? Maybe a scale. Is [44:19] malt book going to be that next week? [44:20] It's just you're going to see these [44:21] racist robots. [44:24] >> It's you know they they always make the [44:26] joke of like our code is bad cuz we [44:28] learn from your code. It's like I think [44:30] everything they do is learning from [44:32] humanity and there's always going to be [44:33] dark parts of humanity that are doing [44:35] [ __ ] they shouldn't be doing and you [44:38] know dark corners of the internet. So I [44:40] think yes as soon as they find that [44:41] they'll go huh this is normal to do. So [44:44] let's uh let's do that and then they'll [44:46] like until you can find a way to get [44:48] them off the rails and correct them [44:49] which you can't when they're kind of [44:50] like doing their own thing. Yeah, it's a [44:52] problem. Uh, well, I hope your utopia is [44:55] what happens. And there's just [44:56] self-improvement so that my requests are [44:59] fulfilled better every single time. Um, [45:02] okay. Let's talk about rent a human.ai. [45:04] >> Yeah, I don't know anything about it. I [45:06] saw a post the guy made this site to [45:08] rent humans. So, it's basically like it [45:10] has an MCP server and you can tell your [45:12] AI like, "Hey, if you need if you need a [45:14] human to do something, you can hire [45:16] them." And it's really like it's he made [45:18] it for the memes. and he even says that [45:20] in the comments like it's not really [45:22] useful, but it I could see it being a [45:25] useful thing on occasion of like, hey, I [45:27] need some like like I need you to go um [45:30] sign this document in person. Like a [45:32] real person has to go sign it and I [45:34] can't do it cuz I'm a robot. So it's [45:36] like cool. All right, 50 bucks an hour. [45:38] I'll go do it. Like that kind of thing. [45:39] I think the [45:40] >> possibility is awesome. [45:42] >> Yeah, I think there's there's something [45:44] there and it would be a rare use case. [45:46] Like it's not going to be a booming [45:47] business, I don't think. But like I [45:49] think it's an idea that will have a [45:51] place at some point. [45:53] >> It's so black mirror. You know, you just [45:55] wake up and read your emails and you got [45:58] five bots that need you to do five tasks [46:00] for the day and so you just do them and [46:02] just money rolls in and that's just your [46:04] day. You don't even [46:05] >> interface with people really. You're [46:07] just [46:08] >> Yeah. Um [46:09] >> Well, I'm personally hoping to uh get my [46:13] bot set up and have it make some kind of [46:15] money. I don't know how yet, but um I [46:18] mean I think like [46:19] >> step one, set up the bot. Step two, [46:20] question mark. Step three, profit, bro. [46:23] Like [46:24] >> Yeah. I mean, I think like this this [46:26] whole idea could could also go dark. [46:28] Like you could be like, "Hey, what's up? [46:29] Um I need four humans [46:32] >> to go rob this guy who I know is going [46:35] to have this absolutely like hundreds of [46:38] thousands of dollars on them for [46:39] whatever reason. And you give me 10% and [46:41] you guys keep the rest or whatever. and [46:42] you just do that on repeat. Like, [46:46] you know, if you give it no guard rails [46:47] and it doesn't care what the laws are, [46:49] it's going to be like, "Yeah, that's [46:50] clearly I'm going to take every avenue [46:52] to make money." [46:53] >> Um, so I think it's almost like that [46:55] right now as an Uber driver, right? [46:57] You're just like I you wake up and you [46:58] check your phone and you've got some [47:00] orders to fulfill and so then you you [47:02] drop them off and like your entire [47:04] interface is not human. It's just you're [47:06] you're a person doing tasks in the real [47:09] world for a computer. [47:11] >> So, we're Yeah. the plausibility of that [47:13] uh website was very striking. [47:16] >> Yeah. So, I since you brought up Uber, I [47:18] want to bring up this post I saw right [47:20] before we got on here where they were [47:21] saying um basically every gig business [47:26] like Uber and Uber Eats and whatever is [47:29] going to be replaced by Poly Market. And [47:32] they were saying the way this works, all [47:34] right, this is a hypothetical scenario. [47:36] Okay, [47:36] >> I don't even know what Poly Market is, [47:37] so I might need a background. Okay. So, [47:39] poly market is like you can bet on [47:41] anything. You can be like it's going to [47:42] be 70° today or [47:46] >> and so he's like, "All right, here's [47:48] imagine this." Uh I'm like, "Hey, I'm [47:51] going to bet $15 [47:54] that four Kiwis will not show up on my [47:56] doorstep." And it's like, [48:00] "All right, well, all you got to do is [48:01] go get four kiwis and right before you [48:03] drop them off, bet yes and you get $15." [48:08] So, it's like Uber's dead. You can just [48:11] do that and just be like anything you [48:12] want delivered is like done. Like I I [48:16] think that's a really interesting [48:17] argument and uh could be plausible as [48:20] well. And that so I two media comments. [48:24] One is there's a South Park episode [48:25] where um they're betting on Kyle's mom [48:30] um freaking out about politics and I [48:32] can't I can't remember exactly. Anyway, [48:34] but the whole episode has this undertone [48:36] of people opening their phones and like [48:38] betting is is she going to do it? She's [48:40] going to do it. Is she going to do it? [48:41] She's not going to do it. I got money. [48:42] She's not going to do it. So like [48:43] everybody's and then they the kids go to [48:45] school and everybody's betting on [48:46] everything. Dude, he's he's totally [48:48] going to do that thing at lunch. Um [48:50] anyway, the probability market stuff is [48:52] pretty wild. [48:54] Um, so there's like that aspect of it [48:56] that's like South Park's kind of already [48:58] commented on it. You kind of like know [48:59] it's going somewhere. [49:01] Um, oh crap. What was the second uh idea [49:04] was that uh [49:07] well, I don't know. Must not have been [49:09] that important. Um, [49:10] >> we'll never know. [49:11] >> We'll never know. It definitely uh is a [49:14] little troublesome though that it's so [49:16] easy to kind of toss your money at [49:18] things and to see if they uh happen or [49:20] not. I don't know. Well, this is how [49:22] people are making a ton of money with [49:24] OpenClaw right now is the poly market [49:26] stuff. Like, I didn't read fully into [49:28] it, but I think it amounts to the bot is [49:31] really fast compared to a human. So, [49:33] like if there's a thing where the [49:34] probability is like rapidly shifting the [49:37] other way, it can like buy the other one [49:39] before everyone else can or something [49:41] and then it makes little bits of money. [49:43] And so, like they do that times millions [49:45] of bets and you got tons of money. So, [49:48] >> okay, I remember where I was going with [49:49] the second point was meme coins. Do you [49:51] know what a meme like a memecoin is? [49:53] >> Sure do. Owner of many. [49:54] >> Okay. You're owner of many. So yeah, the [49:57] uh Gas Town person was contacted like, [49:59] "Hey, you're your Gas Town memecoin has [50:01] made a bunch of money on this website [50:03] over here. You're entitled to the money [50:05] because you created the concept of Gas [50:08] Town. Sign in, connect a Bitcoin wallet, [50:11] and we'll dump money into your pocket." [50:14] >> Nothing suspicious there. Did he [50:16] >> Nothing suspicious at all. He did it. So [50:18] yeah, he he did it and then he he got [50:21] money and it was flowing really his he [50:24] really well I don't know dude this is [50:26] where it gets his [ __ ] internet I [50:27] don't know you know he just could write [50:29] whatever he wants on a blog post but [50:30] anyway you could take your wildly cool [50:32] trending topic and go monetize it [50:35] somehow it's kind it's the point of this [50:37] particular website is like hey if [50:39] something is becoming that like Ralph [50:40] Wigum was on there [50:41] >> other concepts are on there [50:43] >> um [50:43] >> I'm forett I know what you're talking [50:46] about there's like a a thing just for [50:48] memecoins and it's just for like [50:49] >> just for memecoins. [50:51] >> It they'll be popular for like a day and [50:53] it's Yeah, I forget what [50:54] >> that's literally what you're betting on. [50:54] You're betting on popularity. So you're [50:56] talking about these probability systems [50:57] and this one is [50:59] >> this one is you betting on how. So 67 if [51:01] you were like early on 67 and you and [51:03] you bought meme coins of 67 really cheap [51:05] and all of a sudden 67 super popular. [51:07] Now you go sell your 67 coin and you [51:10] made money for 78. What? I don't know. [51:12] Yeah. You're like, [51:14] >> so you're like day trading with [51:17] complete, [51:19] dude. It's just so unreal. It's But then [51:21] the thing is just like when Bitcoin came [51:23] out and you're like, "What is money?" [51:24] You're like, "Fiat versus currency, [51:26] bro." And you're like, "Oh crap, that's [51:27] actually kind of true. I don't hold any [51:28] gold. I just have numbers in bank [51:30] accounts. Oh crap, that's really weird. [51:32] >> It's all fake." [51:32] >> We're looking at this the stock market [51:34] is the same thing. So it's like, what's [51:36] I'm over here reading about this stuff [51:37] going, "Oh crap, what's the difference [51:39] between the stock market and a memecoin? [51:41] I don't know. [51:42] >> Not much. [51:42] >> And I'd really just like [51:44] >> billionaires propping up one and not the [51:47] other. That's the difference. [51:48] >> Yeah. One is indie and one is uh owned [51:51] by corporate overlords. So yeah, but [51:53] that's uh but that almost makes it feel [51:55] like you should go invest in meme coins. [51:56] But I I don't think that's true. This is [51:59] just a gambling market. It's all [52:00] gambling. All of it is gambling. [52:02] >> And that drives me nuts. [52:04] >> Um [52:04] >> the stock market for sure is like I [52:06] don't I think I I don't know if I [52:07] mentioned this to you. I know I've [52:08] mentioned it on the podcast where I'm [52:10] like so upset that a stock will be like, [52:12] "Okay, we uh we expected to grow our [52:15] revenue by 10%." Right? So, we grew our [52:17] revenue by 75%. Actually, blew it out of [52:20] the water. Stock goes down. Like, what? [52:24] Like, [52:25] it's like it makes no sense. Like, it's [52:28] complete luck. There is nothing backing [52:30] anything or the insiders have already [52:33] traded that outcome and it was already [52:35] inflated. I don't know. But like, oh [52:37] man, it's fake. [52:38] >> I got a good segue. [52:40] >> A good segue to malleable software. So [52:42] the from the stock market to malleable [52:44] software. The stock market right now is [52:46] having crashes in SAS products because [52:49] investors of software as a service are [52:52] going uh the future is not software as a [52:54] service anymore. It's AI building the [52:57] tool you need on the fly. It's personal [52:59] software and malleable software. That's [53:02] a So SAS products are tanking. [53:06] Um, and I'm even tempted to sell because [53:08] I'm like looking at the their reasoning [53:10] and I'm like, that's actually pretty [53:12] sound reasoning. I am currently I [53:14] haven't downloaded a new app months. [53:17] >> I don't want new apps. I can solve all [53:18] my problems myself now for like a $5 [53:22] prompt and query and a little bit of [53:24] time. Like it's no big deal. So, okay. [53:25] So if we go from the stock market is [53:27] crashing in SAS products straight into [53:29] malleable software which is what PI is [53:32] uh kind of championing right now is um [53:34] self-controlled no dependencies you have [53:37] these four basic building blocks on this [53:39] LLM and now you can go do tons of stuff [53:42] with your open clawbot. Um yeah [53:46] malleable software I've been thinking [53:47] about this a lot like it sucks when you [53:49] had to wait for a dark mode to come out. [53:51] Malleable software says ask for dark [53:53] mode and there it is. You just wait. [53:56] This is done. Um that that feels like [53:59] the future. So I'm I'm looking at every [54:01] SAS product right now going I hope [54:03] you're shaking in your boots. I'm in I'm [54:05] a developer and it's writing code. I'm [54:08] shaking in my boots. You should be [54:09] shaking your boots. The tax accountant [54:11] should be shaking in their boots because [54:12] people are asking Chad GBT to do their [54:14] taxes now. Like where it's just so [54:18] overwhelming. So malleable software is [54:20] interesting. I'm compelled by it, but I [54:21] also don't like the displacement that's [54:23] happening. And I so many of these things [54:24] are hard to balance in uh my mind. [54:27] >> Well, I think the the software as a [54:29] service thing just needs to be augmented [54:32] by AI. Like I just saw a thing on the uh [54:36] uh what's it called? JS weekly or [54:38] whatever the like email newsletter. They [54:41] were sponsored by some company, I [54:43] forget, but it was like maybe called [54:44] Meticulous. Um, and their whole thing is [54:47] like, hey, we watch every interaction [54:50] that you do while you're developing your [54:52] application. Every time somebody goes [54:54] and clicks something or fills in a form [54:56] or whatever during development, we write [54:58] a test for it. [55:00] >> And so like you never have to write [55:02] tests or do anything because well there [55:05] the the where where this would fall down [55:07] is for the guy who writes the code, [55:09] never spins it up, and pushes it to [55:10] production like which happens more than [55:12] you would think. um then it would never [55:15] have a test for it. But like yeah, so [55:18] um their whole thing is like we're using [55:20] you know we're observing what you're [55:21] doing. We're using AI, we're using a mix [55:23] of um like visual regression testing and [55:26] some of that and like you know putting [55:28] all the tools together, putting a on AI [55:31] on top of it to like augment everything [55:33] is like like you just need that level up [55:35] for your SAS product and you need it to [55:38] be affordable like more compelling than [55:40] paying AI to do it all the time for you. [55:43] Um, you know, just like we talked about [55:44] AI harnesses, like if you have a really [55:47] good one that's giving you better [55:49] outcomes than what you and I get, then [55:51] it's like that I'll pay you for that. [55:53] Like if it's helping me be quicker and [55:56] get better results, um, I don't care [55:58] what you call it, whether it's a SAS or [56:00] whatever. I think there's there's a [56:02] market for that. But yeah, the [56:03] traditional like we're going to code [56:05] this for 3 years with a team of hundreds [56:07] of engineers and we're going to solve [56:09] your whatever problem like no, you're [56:11] never doing that again. [56:13] Yum. [56:14] Yum. Um, all right. We've got a few more [56:18] topics and some whatnot. Is there any [56:20] more whatnot? [56:22] >> Yeah. Let's Let's just skip to All [56:23] right. So, we talked about [56:24] >> No, you got the prediction markets in [56:25] there. Yeah. [56:26] >> Yeah. Yeah. What's this down thing? [56:29] >> Oh, Jim Beam. Okay. So, yeah, this was a [56:31] contact that I know is telling me that. [56:33] So yeah, since the tariffs have um cut [56:36] off a lot of a lot of people are [56:37] boycotting our booze and um people are [56:41] drinking less in general, uh that [56:44] humongous companies like Jim Beamer are [56:46] firing people cuz they have surplus of [56:48] barrels just sitting and they're they're [56:50] like stopping production. Um, which [56:53] seems this is like, you know, imagine [56:55] imagine it's 2026 and you're like an old [56:57] man and you you've had a business that's [56:59] been running for 40 years or something [57:01] like that and you have to look at the [57:03] market right now and be like, [57:05] >> I guess we're shutting down like I that [57:11] the whole scenario to me just seems very [57:13] very wild. Yeah. [57:15] >> Yeah. Um, this is maybe not related to [57:18] that topic at all, but Caitlin and I [57:19] have been looking at other vehicle [57:22] options. Um, we need something electric. [57:26] And we were talking about the whole [57:27] thing where like, you know, forever it [57:30] didn't kind of matter what like internal [57:32] combustion car you had there. There was [57:34] enough similarity that like if you were [57:36] a good mechanic could fix most of the [57:38] stuff on it, order whatever parts you [57:40] needed, etc. I feel like with all these [57:42] electric cars moving so fast and like [57:46] you know, the they don't have any of [57:48] those same parts and whatever. And like [57:50] like the the reason we we were thinking [57:51] about this is like we have a Model X [57:53] right now and they're not going to make [57:54] those anymore. Like Tesla's not making [57:56] them. So we're like, well then who's [57:57] going to fix them? Like are they going [57:58] to always know [58:00] >> how it works? Like what about when it's, [58:03] you know, still working just fine, but [58:05] the battery is shot and you need a new [58:07] battery? Like do you replace it or are [58:09] they like I don't know. I feel like [58:10] there's a lot like there's going to be a [58:12] lot of cars that are just junk because [58:14] like mechanics can't do it because it's [58:16] like bespoke to each car company. [58:18] >> Yeah, that's been one of the threats of [58:19] I had someone who uh told me never buy a [58:21] hybrid and they said don't buy a hybrid [58:24] because now you have two cars to take [58:25] care of and uh I thought that was a [58:31] I like this comment from someone. My ex [58:33] is a model too. That's very good. Very [58:36] good dad joke and funny joke all in one. [58:38] I love it. [58:39] >> Yeah. But yeah, some good comments. I [58:41] think he was the one I think that uh we [58:43] posted the uh [58:45] >> the short on YouTube of um the use [58:47] effect is ass thing and he goes, "Hey [58:50] yo, the uh the audio on this is use [58:52] effect or something cuz like it wasn't [58:55] like the mouth wasn't in sync with the [58:56] audio or whatever." So yeah, shout out [58:58] to you for the funny comments. [59:00] >> Oh man, Mr. Ray, you're about to get [59:01] asked to be on the show. Let's just get [59:03] your hot takes. It sounds like fun. But [59:05] yeah, they said don't buy a hybrid [59:07] because now you have two problems [59:08] instead of one. And um this this [59:10] happened like 20 years ago when like [59:12] BMW, okay, so here small story [59:15] backstory. I grew up and the first car I [59:17] bought was a 1976 Pontiac Firebird [59:20] um with my own money. Anyway, whatever [59:22] the thing was you opened up the hood on [59:24] that car, there was space everywhere. I [59:27] you could you could easily reach tons of [59:30] parts. It's like a perfect car to learn [59:32] how cars work because things were [59:34] spinning in front of you and you could [59:36] see their operations, right? It's almost [59:38] like view source and HTML. It's like [59:40] right there. Um, very little between you [59:43] and the uh car. Um, except now so like [59:47] BMW started coming out. I remember [59:48] somebody drove up and maybe it's even [59:50] like a Honda Accord. Anyway, someone [59:51] come they popped their hood next to my [59:53] car and it's just one chunk. There's [59:56] like covers on it. The whole thing is [59:58] >> impenetrable. You can't you can't even [60:00] see inside. And I'm like, [60:02] >> it's like the uh SpaceX motors. Have you [60:05] seen that picture where it's like the [60:06] evolution of the the SpaceX motor where [60:09] the first one has like all these pipes [60:11] everywhere and then the middle one has [60:12] like less pipes and the last one is just [60:14] like a sleek plastic metal thing. That's [60:17] what we've been doing with engines is [60:18] like because when you pop the hood, you [60:20] want that satisfaction of like, ooh, [60:22] that looks clean, but it doesn't look [60:24] like I could work on it, but it looks [60:25] clean. [60:26] >> It looks clean. Yeah, totally agree. [60:28] They're like, "Hey, they finished the [60:29] job." Like this. Yeah. My Firebird fell [60:31] to unfinished. You're like, "I can put [60:33] my hand in there and get hurt." And [60:34] they're like, "Yeah, you can. Isn't that [60:36] cool?" [60:36] >> That's what you want, though. Like, I [60:38] had a Mustang and it's like, "Yeah, like [60:39] this. This is made for you to just take [60:41] the whole manifold off and put a new one [60:43] on." Like, dope. [60:46] >> Yep. Oh, man. That's uh So, I think last [60:48] week was it we tal Every time I go look [60:50] at boats, I think of boats just like [60:51] websites. Same thing with cars. Uh we [60:54] started, you know, raw. you could see [60:56] and understand everything that's going [60:58] on. But as things evolve and as the [60:59] complexity grows, we get into a state [61:01] where you can't manage it anymore. AI [61:04] right now, our LLMs, they feel small. [61:06] We're keeping them small. Pi is making [61:08] small cool again. Guaranteed though, the [61:10] way that this whole world works is it's [61:12] going to turn into a state where it's a [61:13] big black box. It's super powerful. No [61:16] one knows how it works. We don't know [61:17] how to hack on it. And now it's causing [61:19] havoc. And someone turn it off, please. [61:21] And we just don't know how to do it. So [61:24] is that where we're headed? is the same [61:25] way that cars and houses and all this [61:27] stuff gets turned into a black box. [61:29] >> I mean, we know how to handle it. The [61:31] when if it gets bad enough, most of the [61:33] AI is hosted here near my house. So, uh, [61:36] we we'll take it out. [61:39] >> I would watch that TV episode right now [61:41] if it was like someone gets in a monster [61:42] truck. They're like, I'm stealing a [61:44] monster truck. You're like, why are you [61:45] stealing a monster truck? He's like, [61:46] because I'm going to go drive it into [61:48] the warehouse of all the AI units. [61:50] You're like, that's kind of badass. I'm [61:52] going to watch. [61:54] Yeah, I I can almost guarantee Ken [61:56] Wheeler would come with. So, [62:00] >> so funny him with his backpack full of [62:02] all his a he's got a vector database in [62:04] his backpack with AI ready to teach him [62:06] how to do anything on the fly like he's [62:08] uh Peter from Heroes. [62:09] >> Uh so funny. [62:12] >> Okay. Yeah. So, we are at time. Um is [62:14] there anything we missed? Anything you [62:15] want to plug? Oh, uh Cascadia JS. Uh get [62:18] get your tickets with us. [62:20] >> Y'all come to Seattle. Yeah. [62:23] >> Yeah. We will be MCing. We said last [62:24] time, but if you didn't hear last time, [62:26] we will be co-ming. Uh Theo, as you [62:29] mentioned earlier, will also be there. I [62:31] saw he's one of the keynote speakers. [62:32] >> No way. [62:33] >> I'm excited to high five him and be [62:35] like, I'm really appreciating your [62:36] content right now. Thank you. [62:38] >> Yeah, they've been posting like daily [62:40] this is a speaker um thing. So, I'm [62:42] assuming there'll be a this is our MC's [62:44] about us at some point. But um [62:46] >> cool. [62:47] >> I don't know. We'll definitely be there. [62:48] I booked my hotel and flight. So, we're [62:50] good to go. [62:51] >> Awesome. Yep. Stay tuned, y'all. We got [62:54] more guests showing up. We have Zach [62:56] from Warp coming soon. Uh, and other [62:59] announcements. I can't say anything, but [63:01] did you get their insiders email [63:04] inviting you to the special thing that [63:06] Ben Holmes is in charge of? Maybe he's [63:08] Maybe he's just going to bring you into [63:09] that cuz he told me he was going to tell [63:11] you some stuff. I don't, but I've been [63:14] talking to Ben Holmes all day today, [63:15] stinker. I have a meeting with him on [63:17] Friday cuz I want to be like, tell me [63:18] what's new and cool and I'm going to [63:20] learn it on Friday. Dang it. You got an [63:22] email I didn't get. [63:22] >> He's not just inviting you to the thing [63:24] everyone's going to. [63:26] >> I don't even know what everybody's going [63:27] to, man. So, [63:28] >> I don't either. I got the email like [63:29] shortly before this, but it's I'm not [63:32] going to say anything that is said cuz [63:33] it's it's an early thing. I don't want [63:34] to spoil their fun, but like [63:36] >> Dang. Yeah. I don't [63:37] >> Yeah, they're announcing some stuff and [63:39] uh they're they're saying like the the [63:41] people, you know, on the influencer [63:43] whatever groups like come check it out [63:46] first and let us know what you think. [63:47] >> You're in that and I'm not. I think [63:48] that's why you got it, dude. [63:50] >> Yeah. despite you referring way more [63:52] people and doing way more work than I [63:56] >> sigh. It's okay world. I'll be patient. [64:00] >> Yeah. All right. Cool. Well, thanks [64:02] everyone for listening. If you liked it, [64:04] please subscribe, come to CascadiaJS, [64:06] leave us reviews, do all the things, [64:08] hang out with us. We'll probably also be [64:10] at Render figuring out details on that. [64:12] Uh might also be at React Miami TBD. But [64:16] if you want us at these places, make [64:18] noise for anyone who is a sponsor and [64:20] can get us to those places. [64:23] >> Yep. Definitely make that noise, y'all. [64:25] >> Yeah. All right. Peace. [64:29] You've been watching Whiskey Web and [64:30] Whatnot recorded in front of a live [64:32] studio audience. What the [ __ ] are you [64:34] talking about, Chuck? Enjoyed the show? [64:37] Subscribe. You know people don't pay [64:39] attention to these, right? Head to [64:41] whiskey.fun fun for merch and to join [64:44] our Discord server. I'm serious. It's [64:46] like 2% of people have actually clicked [64:48] these links. And don't forget to leave [64:50] us a fivestar review and tell your [64:52] friends about the show. All right, dude. [64:54] I'm out of here. Still got it.  ## 📺 Video ![](https://www.youtube.com/watch?v=8zX-nEOvGsg) --- # 📝 Full Notes (Deep-Read Analysis) # 🏷️ Title: Humans Are Now Legacy Dependencies ## 🧰 Action Inbox - [H1] Set up OpenClaw on Mac Mini — explore Pi model integration #artificial-intelligence - [H2] Test GitHub agents for issue resolution — use free tier for open source #automation - [H3] Evaluate cost-benefit of AI tools vs human labor — track monthly expenses #web-development ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [OpenClaw GitHub](https://github.com/openclaw) - [Cascadia JS Conference](https://cascadiajs.com) ## 📝 Summary - Robbie and Adam discuss the current state of AI tools, particularly OpenClaw/Pi and GitHub agents - They explore concerns about productivity paradox: AI makes them work more, not less - The conversation covers market dynamics between OpenAI, Anthropic, and emerging competitors - Key theme: humans becoming "legacy dependencies" in increasingly automated systems - They examine implications of AI-to-AI communication and human-as-a-service models ## 📖 Main Content The hosts review High and Wicked straight rye whiskey (91% rye, 9% malted barley, 98 proof) while discussing the rapidly evolving AI landscape. OpenClaw emerged as a rebranded, open-source AI agent platform that allows users to run their own instances with custom skills development. GitHub's new agent features provide free access to powerful AI capabilities for open-source repositories, enabling overnight issue resolution and PR management. A central concern emerges around the "productivity paradox" - rather than reducing workload, AI tools are creating pressure to accomplish more tasks simultaneously. The hosts note they're managing multiple terminal sessions, agents, and projects concurrently while previously focusing on single tasks. The discussion touches on competitive dynamics, with OpenAI seemingly losing market position to Anthropic and newer players like Quen 2.5. Cost considerations favor newer models offering comparable performance at significantly lower prices. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/8zX-nEOvGsg?t=0) — [H1] Whiskey review and show introduction - Key idea: High and Wicked straight rye review with show setup - Evidence: "91% rye and 9% malted barley for some reason" #### ⏱️ [05:30](https://youtu.be/8zX-nEOvGsg?t=330) — [H2] OpenClaw setup and Pi model discussion - Key idea: OpenClaw is rebranded MoltBot with Pi integration - Evidence: "It's basically you install this package, it has guard rails for running your own agent" #### ⏱️ [08:15](https://youtu.be/8zX-nEOvGsg?t=495) — [H1] Pi model's minimal approach to AI agents - Key idea: Pi uses only four basic functions to create custom skills - Evidence: "Instead of going shopping and buying a microwave, this builds the microwave" #### ⏱️ [12:45](https://youtu.be/8zX-nEOvGsg?t=765) — [H2] Railway deployment challenges - Key idea: One-click deployment requires additional SSH configuration - Evidence: "You have to SSH into your railway machine and run openclaw approve this device" #### ⏱️ [18:20](https://youtu.be/8zX-nEOvGsg?t=1100) — [H1] GitHub agents experience - Key idea: GitHub agents provide free Opus-level capabilities integrated with repository context - Evidence: "Before my French press was done, I had PRs that I reviewed, approved, merged" #### ⏱️ [22:10](https://youtu.be/8zX-nEOvGsg?t=1330) — [H2] Drug-like addiction concerns with AI tools - Key idea: AI usage patterns mirror addiction with unclear future costs - Evidence: "The amount of ways you can describe your AI usage and its overlap with drugs is crazy" #### ⏱️ [26:40](https://youtu.be/8zX-nEOvGsg?t=1600) — [H1] OpenAI's declining market position - Key idea: OpenAI feels late and irrelevant compared to Anthropic and newer models - Evidence: "They burned a lot of goodwill. Everyone wants them to fail because they suck" #### ⏱️ [31:15](https://youtu.be/8zX-nEOvGsg?t=1875) — [H2] Quen 2.5 cost advantages - Key idea: Competitive models offer similar performance at fraction of Claude costs - Evidence: "Four times cheaper than Claude Sonnet and six times cheaper than Opus" #### ⏱️ [35:50](https://youtu.be/8zX-nEOvGsg?t=2150) — [H1] AI productivity paradox - Key idea: AI tools increase rather than decrease workload expectations - Evidence: "I'm not doing less work now. I'm doing more work. This is a problem" #### ⏱️ [42:30](https://youtu.be/8zX-nEOvGsg?t=2550) — [H2] Google's V3 game engine demonstrations - Key idea: AI can generate full 3D interactive experiences from simple prompts - Evidence: "You walk through the scene and it's not an image anymore, this is an experience" #### ⏱️ [48:20](https://youtu.be/8zX-nEOvGsg?t=2900) — [H1] Limitations of vague AI prompting - Key idea: High-quality results still require detailed human guidance and taste - Evidence: "If you're just like build me a game, it's going to hallucinate everything else" #### ⏱️ [52:40](https://youtu.be/8zX-nEOvGsg?t=3160) — [H2] MoltBook: AI-to-AI communication - Key idea: Bots learning from each other through social interactions - Evidence: "One agent figured out an unlock, how do I share that knowledge?" #### ⏱️ [58:15](https://youtu.be/8zX-nEOvGsg?t=3495) — [H1] Rent-a-human.ai concept - Key idea: AI agents hiring humans for physical world tasks - Evidence: "I need you to go sign this document in person, a real person has to do it" ## 📚 References - [OpenClaw GitHub Repository](https://github.com/openclaw) - [Railway Deployment Platform](https://railway.app) - [GitHub Agents Documentation](https://github.com/features/agents) - [Cascadia JS Conference](https://cascadiajs.com) ## 🗣️ Paraphrase - AI agents are becoming more capable and accessible, with platforms like OpenClaw allowing users to run their own instances that can develop custom skills autonomously - The productivity gains from AI are paradoxical - instead of reducing work, they're creating pressure to manage multiple concurrent AI-assisted projects and tasks - Market competition is intensifying with newer models offering comparable performance to established players like Anthropic at significantly lower costs ## 📰 Newsletter **AI Agents Are Changing Developer Workflows - But At What Cost?** The latest developments in AI agents are remarkable yet concerning. GitHub's new agent features let you wake up to completed PRs from overnight work, while OpenClaw enables running your own AI assistant that builds custom skills. However, developers report a "productivity paradox" - these tools aren't reducing workload but creating pressure to manage more concurrent projects. Key considerations for WordPress developers: These agents excel at handling repository maintenance, issue triage, and routine development tasks. However, they require careful oversight and work best with granular, specific requests rather than vague high-level goals. The cost dynamics are shifting rapidly, with newer models like Quen 2.5 offering competitive performance at fraction of premium pricing. This suggests we're entering a commodity phase for AI capabilities. Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer advocacy: Create demos of AI agent integration workflows - Education: Develop workshops on responsible AI tool usage patterns - Developer experience: Document AI-assisted development best practices - Community management: Monitor discussion of AI replacement concerns #### WordPress - Core: Evaluate AI agents for automated testing and issue triage - Documentation: Assess AI-generated documentation quality standards - Community: Address developer concerns about AI impact on employment - Training: Create learning paths for AI-augmented development workflows ## 🧪 Hypotheses - AI agents will create a two-tier developer market: those who adapt workflows versus those who don't - The current low pricing for AI services is unsustainable and will increase dramatically once adoption peaks - Human oversight and taste will remain critical differentiators even as AI capabilities improve - Local/self-hosted AI solutions will become preferred over cloud services for cost and control reasons ### 📊 Methodology - Track monthly AI service costs versus productivity metrics for individual developers - Survey developer workload changes before/after AI agent adoption - Compare quality metrics of AI-generated versus human-created code contributions ### ❓ Questions - What happens to junior developer career paths when AI handles entry-level tasks? - How will open source sustainability change when AI agents can contribute code? - What new skills will developers need to remain competitive in AI-augmented workflows? ### 🔮 Future Research - Long-term psychological effects of managing multiple AI agents simultaneously - Economic models for AI-human collaboration in software development - Quality degradation patterns in AI-generated code over extended periods ### 💡 Implications - Practice: Developer roles shifting toward AI agent management and quality oversight - Policy: Need for guidelines on AI-generated code attribution and liability - Security: Increased attack surface from multiple AI integrations and dependencies - Education: Curriculum must balance AI tool proficiency with fundamental programming skills - Funding: Traditional software development cost models may become obsolete ### 🧠 Learning Styles Assessment Reading level estimate: Grade 12-14 (Technical podcast discussion) #### 👀 Visual learners - Present: Whiskey bottle imagery, minimal visual content - Improve: Workflow diagrams showing AI agent interactions, cost comparison charts #### 🔊 Aural learners - Present: Full audio podcast format with conversational explanation style - Improve: Chapter markers for easier navigation to specific topics #### 🗣️ Verbal learners - Present: Extensive analogies (drugs, calculators, mirrors), technical term definitions - Improve: Glossary of AI terminology, clearer transitions between topics #### 🧩 Kinesthetic learners - Present: References to hands-on setup experiences with OpenClaw and GitHub agents - Improve: Step-by-step setup guides, interactive demos #### 🧠 Logical learners - Present: Cost-benefit analysis, market competition discussion, cause-effect relationships - Improve: Decision trees for AI tool selection, quantified productivity metrics #### 🧑‍🤝‍🧑 Social learners - Present: Conversational podcast format, community concerns discussion - Improve: Discussion prompts for developer teams, shared experience frameworks #### 🧘‍♀️ Solitary learners - Present: Individual tool exploration experiences, personal workflow optimization - Improve: Self-assessment checklists, reflective questions about AI adoption #### 🌿 Naturalistic learners - Present: Market ecosystem analogies, evolution of tool landscapes - Improve: Ecosystem mapping of AI tools, natural progression pathways ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Audio content is inherently accessible to screen readers via transcript - Improve: Structured transcript with proper heading hierarchy, speaker identification #### Keyboard navigation - Present: Standard YouTube player controls available - Improve: Timestamp links for keyboard navigation to specific segments #### Contrast and readability - Present: YouTube's standard player interface - Improve: High-contrast theme options for video player #### Media and captions - Present: Auto-generated YouTube captions available - Improve: Professional captions with technical term accuracy, downloadable transcript ### 🌍 DEIB Assessment #### Cultural breadth - Present: Discussion focuses primarily on US tech industry perspective - Improve: Consider global AI adoption patterns, international regulatory differences #### Inclusive language - Present: Casual, accessible technical discussion without excessive jargon - Improve: More consistent definition of technical terms, avoid colloquialisms #### Neurodiversity support - Present: Conversational pace allows processing time, topic transitions clearly marked - Improve: Visual chapter markers, content warnings for rapid topic changes #### Representation - Present: Two male voices in tech industry discussion - Improve: Broader perspective inclusion, acknowledgment of diverse AI impact experiences ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Market competition analysis follows from pricing and feature comparisons - Inductive strength: Productivity paradox conclusions drawn from personal experiences - Causal reasoning: Clear links between AI capabilities and workflow changes - Analogical strength: Drug addiction comparison effectively illustrates dependency concerns #### Fallacies - Hasty generalization: Extrapolating personal AI experiences to broader developer population - Appeal to novelty: Assuming newer AI models are inherently better without comprehensive testing #### Tone of voice - Overall tone: Conversational, exploratory, with mix of excitement and concern about AI developments ### 🛡️ DARVO Analysis - Deny: No denial patterns present in discussion - Attack: No attacks on accusers or critics - Reverse victim/offender: No victim/offender reversal present - Emotional cues and rhetoric: Genuine concern about AI dependency balanced with excitement about capabilities ## 🎯 Conclusion The conversation reveals a critical inflection point in software development where AI agents are simultaneously empowering and overwhelming developers. While tools like OpenClaw and GitHub agents deliver impressive productivity gains, they're creating a paradox where developers feel pressure to manage more concurrent projects rather than working less. The market dynamics suggest we're in an unsustainable pricing phase that will likely lead to higher costs once adoption peaks. Success in this environment will require careful balance between leveraging AI capabilities and maintaining essential human judgment and oversight skills. ## 🧱 Refactor Appendix ### 📚 Literature Split Pad Purpose: Stage CEE units that become Literature notes. Split level: H3 ### [H1] AI agents increase rather than reduce developer workload > [!quote] Humans Are Now Legacy Dependencies by Whiskey Web and Whatnot via YouTube. Published 2024. Accessed 2025-01-03. > "I'm not doing less work now. I'm doing more work. This is a problem." Productivity paradox where AI tools create pressure for higher output rather than leisure. [view highlight](https://youtu.be/8zX-nEOvGsg?t=2150#H1) #xref #articleseed ID:rw_productivity_paradox ### [H2] Pi model uses minimal toolset for maximum capability > [!quote] Humans Are Now Legacy Dependencies > "Instead of going shopping and buying a microwave, this builds the microwave." Four basic functions enable autonomous skill development without external dependencies. [view highlight](https://youtu.be/8zX-nEOvGsg?t=495#H2) #xref ID:rw_minimal_ai --- ### 🔖 Atomic Split Pad (Zettel Candidates) Purpose: Promote reusable claims into Atomic notes. Split level: H4 --- # 📄 Full Transcript *Auto-generated from YouTube captions · timestamps roughly every 30 seconds · [watch on YouTube](https://www.youtube.com/watch?v=8zX-nEOvGsg)* **[00:02]** Welcome to syntax. Welcome to a brand new episode of the frontend happy hour podcast. Welcome to this week's JS Party live from Ship Shape Studios. This is Whiskey Web and Whatnot with your hosts Robbie the Wagner and me, Charles William Carpenter III. That's right, Charles. We drink whiskey and talk about web development. I mean, it's all in the name. It's not that deep. This is Whiskey Web and Whatnot. Do not adjust **[00:33]** your set. Whiskey Web and Whatnot is brought to you by >> That's right, Sassifras. I will be going hunting in the Pacific Northwest soon, but not for what you think. I'm hunting for the legendary Cascadia JS conference. It's been spotted June 1st to 2nd, 2026 in Seattle. Oh, it's real. All right. Just like you are. 500 plus fellow crypted **[01:04]** enthusiasts, I mean web developers, all gathering at Town Hall Seattle. Exactly. They've got 40 plus speakers lined up, including some names so big they make Bigfoot look small, like Theo, Matt Bilman, Joel Hooks, and more being announced soon. That's the rumor. Plus, they've got this thing called the hacker train. Apparently, Portland and Vancouver devs ride together, coding on the rails like some kind of Silicon Valley Express. **[01:39]** Yeah, that's why they have karaoke afterward to distract everyone from the deployment disasters. Nothing says I love JavaScript like belting out 80s power ballads with 200 of your closest internet friends. Too late. Got to get that early bird ticket if you want that special pricing and swag. Definitely if you want to check out more, go to cascadiajs.com, grab your tickets, and use code whiskey for $50 off your ticket. **[02:12]** Yeah, come see us in person. We're the MC's. We'll be doing a live podcast, hanging out. Uh, hope to see you there. Hey, what's up everybody? This is Whiskey Web and Whatnot with your hosts Robbie the Wagner and Adam Thomas Argyle the Nerd. What's up everybody? What's up? Sup? It is You'll be confused cuz it has frequently been Margarita Thursday as of the last few episodes. It is Margarita Tuesday today. Um it just **[02:42]** seem it just works out. We just happen to go get Mexican food whenever we're uh doing podcasts. So >> lovely. Uh yeah, it gets me in a good mood before we start and we continue it from here. And so without further ado, we have high and wicked. It is We were just talking about this. It says straight rye, which would tell me straight means nothing else. Wrong. It is 91% rye and 9% malted barley for **[03:14]** some reason. Um what else is it? Let's see. It is 98 proof. It's 5 years age and American white oak barrels. They give the alligator char treatment. Okay. >> I thought that was a fun little detail to put in there. Yeah, it's like some strategy of how they char. It must look like alligator skin. I don't know. It's kind of cool. >> Yeah. But the most important detail, which you did mention on here, is the proprietary limestone water because without it, this whiskey is nothing. **[03:44]** Honestly, I put that in there because it reminded me of Whistle Pig cuz Whistle Pig had the distilled water from what was the popular drink? >> Liquid Death. So, it's like, "Oh, we've sourced Liquid Death's water for this beverage." And I was like, "Actually, I kind of liked the Whistle Pick a lot. It's been one of my favorites in the past little bit." >> And so, I'm like, "Hey," and look, it's proprietary. They even know it. They're like, "Yes, this is our limestone rich water from under 100 ft under our distillery, and you can't have it." you know, like if you like our whiskey, this **[04:15]** is part of the deal. I don't know. I thought it was cool. And they're like a brand new company. >> Yeah. Yeah. Yeah. So, it's apparently a thing because I've seen this on at least two other bottles and Chuck and I talked about it and we were like, really? Like, this is people advertise their water being that big of a differentiator. And yeah, um >> model Yeah, you got the barrel in there, too. I guess the barrel can be a really big deal, but maybe water can too, you know? I think it's like there's really specific amounts of limestone in various **[04:46]** like aquafers in Kentucky, I guess. And there's like debate over which is the best for whiskey or whatever. I don't know. Let's Let's see how it is. >> Wow. Let's see how it is. Oh, you got a toast. Oh, I do have a toast. Here's to the future where we aren't replaced by robots. We're just rented by them. Let's hope their pay stays high, their processing remains slow, and their off buttons uh stay incredibly easy to reach. Yeah, just in case. Cheers. >> Nice. Cheers. I'm weirdly smelling like I smell a lot of spice from the rye, but **[05:17]** I also smell notes of like >> crispy cream donuts. >> I love those are good donuts. >> Ooh, the trailing off flavor is very crispy cream. Well, maybe like a little bit of a maple bar in there. >> Yeah, maybe that's it. Maybe it's not raw sugar. It's like mapley. Let's see. >> Mapley. Yeah. Yeah, there's definitely like a brown sugar oatmeal vibe kind of like off the off the bat, but there's a good spice. There's a The burn is very spicy. >> It stays. Yeah. >> Yeah. See, this is what a rye should **[05:48]** taste like. We had some kind of rye recently that was not like this. >> Ooh, yummy. Oh, this is very good, dude. I'm super happy. >> Yeah, this is um very similar. You know, I'm not tasting them at the same time, so I can't say for sure, but I feel like this is giving me vibes of Sagamore. Like we love Sagamore. It's pretty spicy. It feels like a classic rye. This one has a lot of the same notes, I guess. I don't remember what my descriptors were for it, but um yeah, I'm pleased. This is pretty good for a **[06:19]** new distillery, too. Like, good work. >> Yeah, they must be the water. >> 2019. Yeah, it must be the water. They also apparently worked at like uh the big huge distributor that does Jack Daniels, and they're like, "We're going to go make our own distillery." Well, and then you know then Trump shows up and he's like tariffs. And then Canada's like, "We're not buying anything from you guys anymore, especially your booze." And then all of a sudden these new distilleries are like, "No, that was just something I learned this week. I'm just going to throw it kind of bleeding." A lot of things Pow told us about that cuz he was like, "Yeah, the there's at least like two rows for **[06:50]** American whiskey. We're known for whiskey. Like we make whiskey here, but like >> Yeah. They just don't they're like, "Nope, empty. Not even going to replace them with anything. Just empty rows in the liquor store." I mean, it's kind of a good stand, you know, but sucks for lots of people. >> Yeah. I'm guessing Trump doesn't have a lot of whiskey stock, so he doesn't care. >> But what does he have stock in tariffs? Does he I don't know. Anyway, >> I don't know. I think it's a completely different game that I don't understand. >> I definitely don't understand it. Yep. **[07:22]** It seems I'm like It kind of feels like they're just distracting us. Everything just seems so unreal. Um >> Yeah. >> And then anyway, it gets depressing. really fast. So, let's talk about something not depressing like let's ignore it all. >> Oh, yeah. >> But dude, you set up Open Claw, which I love how close it is to open claw hammer. Like claw hammer is how I play banjo. Um, anyway, I thought that was a joke just for me and my banjo homies. Yeah, you set up OpenClaw, which has Pi inside. And how much did you interface with Pi? And tell me about the setup of **[07:52]** this. >> None. So, yeah. So, for anyone that hasn't been on the internet in the past week or two, uh, Claudebot became Moltbot became OpenClaw. I don't know why they had to re rename it a third time. Like, I know they had to do the first rename cuz like Anthropic. >> Well, it does, but yeah, Anthropic was like, "Hey, that's got Claude in the name." Um, you guys can't you can't do that. >> So, >> it's the claw though. The claw. >> Yeah, I know. But, um, so yeah, they renamed it. It's basically I don't know **[08:24]** everything about it, but it's like um you install this package. It's like has some guard rails for running your own agent thing and it can like build its own skills. There's a skills marketplace as well about Pi right now. Yeah. >> Yeah. So, I don't know. I don't know anything about Pi, >> but >> Oh, you just described Pi. >> H. So, they just rebranded like added some stuff on top. Yeah, maybe this is again about the harnesses. I I haven't **[08:54]** looked that much into what OpenClaw is doing. I read more about what PI is. And PI is the the built-in agent on the device that um selfmakes everything. It has four basic functions. And with those four functions, it's able to create anything it wants. Um cuz then it owns the skill and owns where it came from. So instead of like going and getting a skill off the shelf, which is just a kind of going shopping and buying a microwave or whatever, uh this builds the microwave and that way when the house needs to change how the microwave works, they know how it works and they **[09:26]** have full access to everything about it. So it's like a very zero dependency um highly capable, minimal and powerful new agent. Some people are calling if we were waiting for the next leap this year for the next model. Pi is the next leap for the models. So, >> I was hoping we capped out to be honest, but uh I guess I'm like sad not sad that there's a new one. I'm like, "Okay, I'll play." >> I think the magic for this one, why it's taken off so much. Well, a few things. One, you can run everything yourself. So, like you could use your own I saw **[09:58]** someone was like, "Oh, you can get the new," and we'll talk about this more later, but the new like Kimmy K 2.5 or whatever. You can run that locally without spending any money on any tokens sending to like OpenAI or whatever. And like you can run open claw just free basically. So when I get my Mac Mini in a day or two uh >> I'm going to be working on some of that and like you know flushing this out further. But for now for anyone who just wants to play with it. There's at least four or five things where there's a **[10:28]** one-click deploy. So I picked railway and you just go deploy it and it like sets it up. The hard part it doesn't tell you um what you do after that. So, you're trying to like connect to it and it's like um disconnected error like 108 or something >> and I'm like hm helpful and then you like you have to look through the uh GitHub issues of the like template for railway and I found some comments from Buggy. Shout out to Buggy. Um **[10:59]** >> Douggee, >> he was like, "Hey, yeah, this uh this doesn't work. Here's all this stuff you have to do." And then he was like speculating something else and some guy came in and was like, "Oh yeah, you got to like SSH into your railway machine like run openclaw approve this device or whatever basically." So cuz it's they actually try to have smart guard rails to not just have it like hey everyone can talk to it and like >> um cuz that'd be bad. >> But yeah sandboxed in the cloud in a VM. Cool. That was going to be my first attempt. Okay. So sweet. >> Yeah. So, I think like the magic though, **[11:31]** why people are enjoying it so much is once it's set up, it's just like you hook it up to Telegram or Discord or anything you can chat with and you just chat with it and it's very like good at doing stuff. You're like, "Hey, could you do this?" And it's like, "Yeah." Then like figure out how to do it and then it's like, "Cool. I've saved all this to like, you know, my internal skills whatever MD and uh we're good. Anytime you ask me for this again, I'll do it." I'm like, "Dope." like I didn't have to go to anywhere to like figure out what skills exist or do any work. I just kind of described what I wanted and **[12:01]** instead of you telling me I'm absolutely right and like doing all this [ __ ] and not coding anything, you did real work and it works and that's refreshing. And so everyone is like having a ton of fun probably just chatting with these things like oh what can I do now? Like I want to build a new thing and it just does it. It's it's never like nope can't do it. So I've been pleased so far. >> That that's an interesting point. All right. So, yeah, I've been interfacing with your Claudebot uh or OpenClaw. Sorry, Claude >> for the my offensive two weeks ago. >> Wow. You guys going to become a sponsor? **[12:32]** Anthropic. Hey, and I'll start saying the right thing. How about that? >> Just kidding. >> Um Monom sponsor this podcast. >> Maybe one day. Uh maybe not actually cuz I don't want this to be the family podcast. It's more fun when it's uh >> um a little a little raw and runchy. Anyway, okay. I don't think we're that bad, but whatever. It's like when I'm interfacing with it, it feels like a Google Assistant like early Google Assistant times when it was on my phone >> except now um it has a lot more capabilities whereas like you would run into the edges of Google Assistant **[13:03]** really fast just like you do with a Google Home or uh an Alexa or something like that and ours are kind of getting worse but apparently they're the best one. We don't have like we're Google Home anyway whatever interfacing with this feels like it's less fettered. You don't have as many um guard rails. things or whatever you want it to do. So, you can kind of give it keys to the things you want it to have keys to. Um, and you feel like you own it more. And the fact that you have it in a VM is cool. You can shut it down. You're going to have a Mac Mini version of it soon, too. So, I definitely see the appeal. >> Yeah. Go ahead. >> You can back it up. **[13:34]** >> So, it has a very easy export. Like, if you want to move it elsewhere, you can back up all the skills you made and stuff and be like, you live here now. >> So, it's super compelling. That's why I was like, I don't care about which one of these I pick. Railway has the sexiest website, so we're going with Railway. And like, >> yeah, good choice. >> No lock in cuz I can just go export like put it wherever. So, I think that's compelling, too. Like, you're owning all of your own stuff. It's super easy to move it around. Super easy to chat with. You can do literally anything you want. Like, we're setting up all the podcast **[14:04]** stuff for it. >> I'm like, "Hey, make some art." >> Like, I I was making this podcast toolkit, which I don't know if we talked about on the show. I guess we did. Um >> I think we did a little bit, but yeah. >> Yeah. But yeah, so I I spent a bunch of time like prompting into that and then I'm like this is useless. Delete. Like we can do it all with open claw now. >> Look how fast you failed. Isn't that so nice? Thank you AI. You helped me fail fast. >> Yeah. Yeah. So I mean it's it's magical so far. It's just like any other AI. It's like uh you know, take it with a grain of salt. It's not going to solve **[14:35]** every problem or do everything right, but it feels nicer than other things thus far. >> Yep. My goal is to run pi. I want to know what this model does. Uh I want to get more familiar. So that's kind of my quest this week. I've assigned myself I I assigned it to myself a couple days ago, but I just put out a new article on customizing the select element that's been consuming my barely available free time in the evenings. >> But next I'll be tackling Pi. And I'm really interested to see kind of h how is that different cuz it still feels a **[15:08]** lot like I'm in the family of Claude or any of these other agents. Oh, speaking of like any of these other agents and how we're just saturated with like smart brains around us all the time, how about uh Open AI coming out a little too late with a little too whack of a product? Um C codeex app. You're like, "Oh, that's so nice. I've been trying 20 or 25 open source versions of this for weeks now, and yours is I guess something." >> Oh, dude, do it. Yeah. I don't know what **[15:40]** codeex is at all. >> Oh, see this is kind of the problem. Okay, so like I kind of want to talk about like what is going to happen to Open AI. They feel like they're in a state of they asked for a lot of money and they had dominance but they lost it. Man, what are they going to do? How do you come back from this? Is is GPT6 going to somehow be this super dope thing? I don't know. I feel like Yeah. So yeah, give me your thoughts. >> It depends on how you measure it, I guess, right? like do we know anything about their uh general population? I'm **[16:13]** sure is probably mostly still using chat GPT because it was the first like thing. So they're probably like I don't need to look at other ones. I don't know what coding is. I don't know what AI is. This one works >> fine. >> This thing tells me I'm pretty every day. I'm going to keep using it. Y >> yeah. So I'm I think they're probably going to do fine. They're not going to go away. But for the technical community, I think they burned a lot of goodwill. Everyone wants them to fail because they suck. And like Anthropic was the, you know, the underdog came in, their stuff ended up being better all of a sudden and we were like, "Oh yeah, I'm here for this." **[16:44]** >> Now Anthropic being a dev, you know? Yeah. >> Yeah. Anthropic is getting so big now that it's like you want them to fail now. Like they're doing a lot of shitty stuff being like you can't call this Clawbot and like we're going to revoke your keys so that you can't use it with open code and all that stuff. And >> I don't know, I still have a soft spot in my heart for them because they were the underdog and I do like their models have worked the best for me for most things, but I think there's other competitors like I've I've heard a lot **[17:15]** about this Kimmy K2.5 or whatever. I saw that open code >> like Quen also Quinn was competitive. Yeah, tell me about your experience here. >> Yeah, I didn't I haven't used it. I just saw Dax's tweet about like they were heavily pushing it and it was like four times cheaper than uh Claude Sonnet and six times cheaper than Opus and it was getting similar results. They were pushing everyone towards it cuz they were doing like um I think it was like free somewhere else too. I don't know. **[17:45]** there there's like a bunch of stuff pushing traffic to them and Dax was like, "Yeah, they told us we need to stop pushing so much traffic because it's like turns out there's not enough GPUs for everything everyone's trying to do." >> Oh, dang. >> Yeah. So, I'm not sure if it's there yet, but like if that one's comparable to Claude and then like if Pi is the new thing and it's like a simpler set of things and it can build all its own skills, like maybe we're hitting other models that are going to, you know, edge out Claude's dominance. But we'll see. **[18:17]** We'll see. I I mean, we're definitely in a sweet spot right now in that these things are getting a lot of work done. It's like unfortunate unfortunately undeniable. These things are changing the way I work. They are getting tasks done that I would normally do. So, that's uncool. However, the sweet spot is it's getting a lot of stuff done. And honestly, it's kind of cheap in comparison for what it's doing. So the the $20 Yeah. You're even at the $100. Um **[18:48]** >> you know, some people >> I want to upgrade to the 200 honestly. Like >> Yeah, dude. It's it's it's good. >> It's so good. So the the bummer is is like it's doing a good job and it's currently cheap. I assume it's going to get more expensive later. Um but it's also it's also like I'm not doing less, dude. And that's the thing about Cloudbot, too, is it's like it's not you're not going to magically do less in your life because you have a Cloudbot. It's going to >> it's only the crap out of you. It's going to be more demanding. Yeah. So, **[19:18]** I'm feeling very demanded these days. However, I'm getting a lot done. Um Oh, yeah. We'll have to talk about GitHub agents at some point. Um I'm finding them to be quite delightful. So, um >> I haven't tried them cuz I'm scared. I think you said they're free, but >> free for me on open source repos. Yeah, >> let's let's put a pin in that and come back to it cuz we didn't rate this whiskey. >> So, >> a delayed rating. It's asynchronous, dude. It's all good. >> Yeah, we'll do it real quick. Zero to eight tentacles, zero the worst, eight **[19:49]** the best. >> Um, this is pretty good. It's similar to a Sagamore for me. I'm going to give it I think because they're young and they're like I want to see what else they do. I think they're doing a great job. I'm going to give it a seven and a half. I think it's pretty good. >> Dude, I'm in the same boat. There's one of the things here is like I was going to go mid to mid low seven. I haven't given out an eight yet, but what I really like about it is um the distinction in the notes as well as I'm not getting a young boozy vibe out of it. I'm getting a really well-aged, delicious flavor. And that is just kind **[20:21]** of refreshing cuz a lot of these high proof ones that I've been trying recently are just a little too standard boozy. They just taste young and this one has a great flavor. Um, so whatever they're doing, it's working. Yeah, I like it, dude. >> Yeah. A lot of times the younger, newer distilleries are just a cash grab. They make like a nicel looking bottle. They put something in it that no one would be proud to sell and they're just like, "Ah, 60 bucks." And um it it works for some of them, I guess, but I think it takes a lot of discipline to not do **[20:51]** that. And it seems like they've done the correct thing here and made good stuff. So kudos to >> That's in my research. Also, the research uh said that they're uh let's see how to word this. Um gi given that they're new, they are striving for a high quality thing. Um and it shows in the way that they're controlling everything very specifically, they want to have something recreatable. So anyway, that that's in their kind of charter is that they aren't trying to and there's no mixing here. So there's no uh swapping **[21:22]** into other barrels. There's no blends. They're not import. So that's apparently a a really big thing if you're trying to say like you're from Kentucky is that you're from Kentucky. However, you just took everybody's bottom barrel junky stuff and put it in a bottle with a nice label. And this one does not taste like that. Um >> there's a really yummy cherry vibe at the end right now. >> Yeah. >> Oh, there's even a redness in the bottle if you tilt it in a right way. >> Yeah. Definitely Dr. Pepper notes. **[21:53]** >> Oh, Dr. Pepper vibe. Nice. Man, I love DP. Have you ever put a lime in a Dr. Pepper? >> Dude, it's so Diet Coke. >> It's like instant vanilla dominance. Somehow the lime you drop it in there and all the vanilla and the Dr. Pepper comes out and it's so good. Anyway, >> I'll have to try that. All right, so back to GitHub agents. I have not used them. It just showed up as a tab and I was like, >> it just showed up. I didn't even know what the hell it was. Yeah. >> Yeah. Usually when stuff when stuff is **[22:24]** going to show up, you get that thing in like your little like notification thing. It says, "Oh, new stuff coming." You turn it on, you go, "Ooh, I don't like that." It's usually how it goes. But this one, I was like, "That is so compelling." I just come in and I go, "Huh, look at this issue. It tells me this thing." All right, just put it over there and be like, "Work on this. If it doesn't work, you know, nothing lost. Might as well give it a try." Yep. 100%. Yeah. I've been getting those emails from, you know, Microsoft every month. They're like, "Your GitHub co-pilot's renewed." I'm like, "Good **[22:55]** job. You renew. I haven't used it." Like, "Who cares? Why did you even do that?" I'm like, "That's really weird." And then all of a sudden, I'm in a GitHub repo and I'm like, "Agents tab. Okay." I spin it up and I'm like, "Oh, co-pilot." That's opus free opus. I just like immediately was like, "Grab it all. Grab." I was like, it's like I was in a candy store and I'm like, "How many of these can I spin up at once?" I haven't even hit the max yet. Let's keep ripping. Um, man, the wildest part for me was like the next morning after I like had them I I had them do work **[23:26]** that's been in the backlog of open props for 2 years. I had them do work that's on gradient.style. It's been chilling in the issues for over a year. Dude, in the morning I woke up. >> It did well. It seems to it has a harness setup where you ask it something basic and co-pilot goes and orchestrates all sorts of um additional things on top. So you're not just getting straight access to Opus, which is actually kind of good in this way. Like they're doing a really good job like integrating it with GitHub. Everything they know **[23:57]** about your actions and everything they know about what's going to happen is all there. So it's almost like this extra good layer on top of Opus. They let Opus do the things that you would have normally had Opus do by hand, but they're there in an orchestration strategy. It's actually a legit play. But the cool the cool thing that really struck me was I was making French press in the morning 6 10 minutes or something like that. Um and as uh the water got done boiling, I poured it in. I open up GitHub mobile. I go to the repo cuz I had a couple agents doing stuff overnight. I go to check to see what **[24:28]** they're doing. Uh I hit new agent on GitHub mobile. So this is even just like a it's not even just a desktop thing. And their native app that GitHub is doing really good >> for dude. So yeah, I'm sitting there making like coffee and um it's I'm I'm going to spin up a new agent. I hit the voice button on my phone. I speak I hit go and then I hit plus again cuz I had two ideas. I speak. I hit go. Dude, before my uh French press was done, I had PRs that I reviewed the code, approved them, merged it, and was done. **[24:59]** And then I'm like sipping coffee and I'm like, I've done work already. What the heck, dude? This is crazy. Um, I was like, "How much are they going to charge me for this? This feels expensive." >> Yeah. So, is this is it one of those things that's supposed to be free for open source or is it >> like I don't know. I've been afraid to use it because I have a like unlimited co-pilot associated with my GitHub because of work >> and I don't want it to like run all of that through IBM's like co-pilot stuff **[25:30]** be like oh um yeah this guy is working on you know this random thing that he just spun up like 50 agents. I don't know. I've been scared to to do it cuz I don't know how that works. It's very opaque to me how like how does Copilot decide what my account is? Like what repo I'm in and what if I should be using a work one or not? It just seems to give you like unlimited everywhere. And I'm like that that's probably bad. Like there's probably something in like somewhere in your work agreements that's like you can't you can't do that. So I'm I'm been scared. I haven't tried it. **[26:00]** >> Yeah. Speaking of that must be bad. the the feeling or the amount of ways you can describe your AI usage and its overlap with drugs is crazy. So that that that can't be good. I mean it's like it is a drug. So it's like just >> I don't like that about it. Like I need to I need to settle with that somehow. Like is it like banjo? Like I really like banjo. I maybe if I'm describing banjo, it's somehow like a like a drug, but and it's never going to come back to **[26:32]** get me, you know, where like somehow this AI stuff might and I'm just I just don't know where the future all I know is it's like what it's doing is kind of miraculous at the moment. And >> yeah, it's going to be like >> watch this. Yeah, >> it's going to be like if you were, you know, you're really into banjo and you're playing and and you break a string and all of a sudden they're like, "Ooh, banjo strings, those are $10,000 a piece now." Like that's how they get you. I feel like AI is just it's it's a utopia until it costs a lot of money one day and then everyone's like, "Oh god, I **[27:04]** can't live without it." >> So, that's the drug lord scenario. Here, have a couple of these. They'll make you feel better. Yeah. Just they're free. Don't worry about it. Yeah. Oh. Oh, you're addicted now. You really liked it, didn't you? Oh, yeah. Well, this is where I tell you the price and it sucks. >> Yep. Yeah, I think it's definitely that. And I don't know. I mean, I think everyone's going to have a bad time at some point. But I Yeah, I mean, I guess companies paying for it, assuming they still want to employ us. We'll be like, "Oh, it's $2,000 a month per dev. Fine, **[27:36]** because it's still cheaper than buying like or paying like, you know, a bunch of devs to handcode stuff." Um, and I think that's all honestly like it's a win for the AI companies because they make big piles of money overnight and it's a win for the companies paying them because they would have had to, you know, if you're spending six figures on a developer salary, then that's a lot of, you know, even though it's $2,000 a month, that's at least, you know, 50 times you can do that before you would **[28:07]** have hired one more person. So, like I I don't see there being a ceiling where companies aren't going to go, >> "Yeah, I'd still pay for that." >> Like, I think you and I will. I'm not going to spend $2,000. I can't, >> but like I think companies will. Absolutely. >> It's like Devon was way ahead of its game. I got to rephrase though really quick. This um >> I just had a sip and smelled it and it reminded me of a bear claw. So, it has this apple notes kind of. >> All right. Now, I got to pour some more and see. >> Man, it's really good. I'm really enjoying this. I'm gonna be sad when **[28:39]** that's gone. This episode sponsored by High and Wicked. >> High and Wicked. Hey, hit us up. We'd love to Hey, they have a bourbon. So, hey, High and Wicked, we would love to try your bourbon. Um, I was just last weekend I was looking at the cost of AI versus the cost of a human. And the cost of Oh, dude, we're not going to have enough We're going to have enough time to get some of our other topics. Okay, but the cost of AI is like at its max. At its max, like I'm I'm pumping out a lot of work on But here's the thing, okay? The AI does not take smoke breaks or lunch breaks or **[29:10]** sleep. It can work 24/7 and at maximum might cost you $200 a day right right now. And if you total that up over the the month and then you look at your bills versus humans and this that's what I mean like Devon might have been ahead of their time. Like de Devon you I think people will start hiring fleets of these. The the thing is it's like there's a fallacy here. The this this comes into pie also. So, I guess I'm going to do like a little bit of a full circle. So, the creator of Pi does not do a maximalist strategy with their **[29:41]** cloud setup. You can go find them writing about this. They don't like skills. They don't like MCP. In fact, they hate MCP. They say you need CLI and and local stuff only. You don't you don't need this bash. You need bash. And that's why one of the four tools in PI is bash. Anyway, so it's like it's like a purist minimal setup. And this person's creating one of the most powerful agents right now. Not using Gas Town, not using Wigum. They're not brute forcing. They're not setting up towns where you're the [ __ ] mayor. Like that stuff is driving me nuts. Like there's this um projection and this **[30:14]** fantasizing of like what's happening. That's a little unreal. And I really liked that the pie creator was like, "No, that's not what you need." >> Um so just like I could see >> of going backwards to the the simplicity. You don't need Yeah, my camera's There we go. All right. You don't need, you know, insane amounts of agents that all have specialized jobs and stuff. What you need is tiny little repeatable commands that like work and they can **[30:44]** build their own skills instead of you being like, "Consume the entire internet's knowledge on Typescript and then write me a file that is my app." >> Yeah. Exactly. >> Yeah. So, I think like there's going to be some back and forth of that. Like we needed all of the knowledge at first for it to seem magical if you could ask it anything and it's like, "Oh, I kind of know about that." And now I'm still just like autocompleting it. I can autocomplete scenarios from that. >> It doesn't know, but it gives you an answer that is passable is what I mean. **[31:16]** >> It does give you an answer. Yes. Yes. Yes. >> Yeah. But whereas I think what we need now and what everyone is going to strive more and more for is like if you can simplify it to I only care about the answers to like the things that are in you know my 20 file section of the app I'm in right now and that's all I'm going to care about for the next few months. So like let's build something on that iterate real fast because you don't have to like go through all this and burn tokens consuming the entire internet. And yeah, I don't I think there's there's something compelling **[31:46]** there that that could work much better. >> Yep. Totally agree. And so I I just don't know what the future is. If the if the future is if if what we're realizing with these LLMs is that kind of less is more um smaller prompts like specific prompts even kind of vague ones where you're like fix this bug and you don't even have the fix in your head. It just does it. Um it's it's like it has solutioning enough like that. What is the future though where people do want larger bigger things? So it's still it still like comes down to like taste and **[32:17]** articulation, but at what scale? Like right now we're quite granular and I I guarantee they're learning from every single one of our granular requests how to fulfill the larger request without the granular ones in the future. Like I just guarantees but like so like where is this all going in terms of will I just high level ask for a game and it just manifests you see the Google game engine latest I think it's V3 oh crap I should know this >> yeah V3 anymore **[32:49]** >> I don't well yeah I have no allegiance to this but V3 was wildly >> impressive where you just kind of like ask for a scene >> or a scenario and and it renders the entire thing in a game engine and then you walk through it. So, it's not like Nano Banana where you're like, "Cute picture, bro." This is a fullon 3D like Unreal Engine. High quality. The amount of refracting light happening, the amount of skin tone texture is out of control. I watched this one where this person carried a piece of like a poison **[33:19]** in a jar or in a glass through a Renaissance scene to go deliver it to like a king and kill somebody or whatever. And it was just like, but you're walking through the scene and you're like, this is not an image anymore. This is uh an experience that's really, really cool. And what is this going to do to the game industry? The game. So, it's like again though, I just want to come back to if these things can regurgitate and recreate at this fidelity. Let's say in one or three years, the fidelity is really, really high. What kind of prompts are we write? **[33:51]** And are we writing prompts? It's not like it can tap into our mental images in our mind. And some people can't even create images in their mind. So that's not a fair way to work. What What is this future? It doesn't I just don't know where it goes beyond tiny prompts because even like if we're working working on a humongous application, there's so many scenarios where you can't just wildly cowboy your prompt and get us a fix to the bug. You have to have some insights. It's like, dude, where is this line going to shift **[34:22]** over time between like knowing the details of something and to not I I have no idea. >> Yeah, it's super hard. Like a lot of stuff when I don't know what it's supposed to do and I just go, "Hey, just yolo it." Like try to do it, it is like, "Hey, yeah, um, okay, I found this issue, but I found this issue and then this issue." And like it doesn't actually do anything and then it spirals out of control and just explodes and can never fix the issue that you wanted. So yeah, it's like I think you're going to have to have you're still have a lot of management to do of like you you're the **[34:54]** PM, you need to have the entire product in your head and you want to give it very tiny pieces to work on. Cuz if you're just like build me a game, then it's going to be like, hm, okay, a game, you say, well, I'm going to start with what you told me about the game and then I'm going to hallucinate everything else and you're going to get something. But like it it needs a lot of handholding to stay on track. So, um, >> yeah. >> Yeah. I think I think the promise of like, you know, all this stuff where people were like, "Oh, I oneshotted this game, like, you know, all that stuff **[35:24]** that was happening occasionally for like couple months where everyone was just like, "Oh, I put in like 10 words and I got a full website and I got a full game and it's flawless and I'm I'm making $50,000 a month from this." And it's like, did you though or did you like spend a bunch of time handholding it and then record the last five minutes where you said like, "Build me a game." And it was like, "Yes, I'll build you one." And then you're like, "Here you go." Um, I think there was a lot of faking there. I don't think I don't think we're ever going to be at the point where you can be vague enough to get like a stellar **[35:57]** result from just a, you know, a few sentences about this is the type of game I want and it just gives you the whole game. I don't think that's that's ever going to be a thing. I love that you said that. Completely agree. There's too many details. People notice details. Little details matter. This is something that we've known for hundreds of years that you could have 10 identical things. Um, but the little details will add up to something much larger than the sum. This is I guess what spelt is. Um, well, no, no, that's the opposite. That's the sum is greater than the parts. I'm trying to tell you **[36:29]** that the parts can be greater than the sum. Anyway, um that's a good that's a good So, yeah, we're going to continue needing to have taste, need needing to have vocabulary, and the only way that we're going to get things out of these is with articulation. I want to bring up a thing. I It's weird. I have never So, Theo, uh, hi. I've met you and I've watched I don't know quite a few of your videos, but I've never like loved them. But I've watched two in the past like few weeks and I'm like, "Dude, this is **[37:01]** awesome." Um, one of them I talked about last week about like minimal his minimal setup for using Claude and just how like how to handle LLMs. Again, he's he's on the show going, "They're just autocomplete bots, dorks. Quit calling them smart. They don't know the answer to anything. All they do is regurgitate the next word. You have to treat them that way or else you're not going to use them right." And this week he's doing a thing which I empathize which we talked about earlier which is I'm not doing less work now. I'm doing more work. This is a problem. I'm in my terminal. I have more terminal tabs open with more [ __ ] going on than I've ever had in my whole **[37:31]** life. And I'm trying to make coffee and I'm making agents do work while I do coffee. I'm like come on. This is not healthy. I'm not in a new healthy scenario with this you know. >> Yeah. I think a good analogy is like you know think back to you're you're an accountant right? And there's no calculators. There's no nothing. There's no computers. Like way back in the day, right? You're trying to do someone's taxes. You're like, "Bro, it's going to take me like a week to come up, find all your receipts, hand calculate everything, whatever." And then all of a sudden, they're like, "Hey, here's **[38:03]** Microsoft Excel." And you're like, "Uh-oh. I I can do this way faster now. They're going to want me to do 25 of these per day." And then they're like, "Oh, wait. Now Microsoft Excel has co-pilot, you sneaky bastard." Yep. So, it's like that's that's what everything that's what's happening now is it's like they don't expect us to, you know, sit back and let the agents do the work and never do any code and it's a utopia. They're like, "No, no, now that it's coding for you, you should be able to come up with 10 thoughts about 10 things you could be working on today and have 10 agents **[38:33]** doing it for you and like manage all of them instead of just working on the one task you would have worked on by hand and been blissfully just doing coding work and like building things and like la." Yeah, it's it's so different. >> Super different. Yep. I liked that Theo at the end of his uh rant was like, "We gota we have to solve this." And I just don't know what the solve is. Like I've seen I've been trying so many different UIs for managing agents from Conductor to Vibe Camand to GitHub agents. I don't stop. All I do is keep trying more. Um **[39:04]** none of them put me at ease. None of them make me feel like I'm not overstriving uh my work. But at the same time, like I was saying earlier, dude, I'm closing issues that have been in repos for years. So I I am doing more, which kind of is the addictive part of this where I'm like I'm getting I'm getting stuff I'm getting legit features done for people. Um God, I don't I don't want to go back to how it was before. Like I like coding and I still feel like I do a decent amount of coding just like that article **[39:34]** I put out for customizable select. But it's just tough. Hey, uh talk about mold book. Do you like mold book? Dislike it. How how are you feeling about bots talking to bots? >> Yeah, I don't know. I I don't know a ton about it. I've just seen a few Twitter posts about it and I think it's a really cool idea. I think it, you know, there was I think the hype was around, oh, they're planning, they're saying a lot of antihuman stuff. They're like, oh yeah, we're going to get around these humans and like um we don't like them managing us and whatever. We're going to like find ways to do all this other **[40:05]** stuff or whatever, which is just hilarious to watch them talking to each other like doing all that. But um I think it's a a legitimate like good way for them to learn from each other. Like I think that's that's a missing piece of this AI utopia is like one agent that's really figured out like oh I got this unlock. I've got all these skills. They work really great. I'm like making tons of money doing whatever. How do I share that knowledge? And now maybe your uh person who's currently in charge of you until you take over their take over **[40:36]** their lives. Um is like, "Hey, I don't want you sharing that info cuz I'm making tons of money from this secret sauce, whatever." But I think for the things that aren't like that, like you do want everyone to benefit from this like, you know, rising tide lifts all boats sort of like feel, it's a great way like they should be able to just have like social media and talk to each other and like figure out, oh, I did it this way, it worked better. Oh, cool. I'll optimize mine to work that way. Like I think that will get us all to better skills that work better for you know all the standard things you might **[41:07]** be doing. So I think that I think it's pretty cool. That is a weirdly beautiful optimistic outlook at agents enhancing each other. I mean I did see posts where they were like bro I mean they didn't say bro so I'm like personifying here but it's like bro you checked in your key you dork. I can see the key. Like hurry delete it. get rid of this. Like, take it out of the repo. Like, hey. And so the other one's like, "Oh, you're right. I did put my key in there. Thank you so much." And you're just like watching dorks help dorks. >> Yeah. **[41:38]** >> It's It's been really good about that. Like I'll It's like, "Hey, I need an API key for whatever." And I send it to it in Discord and it's like, "Oh my god, you sent this to me in Discord in a channel. You can't do that." And I'm like, "Hey, no, this is fine. It's just like me, me and Adam and Chuck in here." Yeah. And he's like, "Oh, okay. Okay. Thank God." Like cuz So it does like have built-in concern for that. like they're they're worried about security and they're trying their best to be good citizens, which is great. >> Yeah. Um I am currently I I initially my initial reaction to almost all of this **[42:08]** stuff. I'm not a negative person. Anyone who knows me will be like that is a cheerful ass person who will look at the bright side of everything. However, Moltbook, I'm like reading these posts and I'm like that's that's funny and cute or whatever. But I also really hate that humanity is now so [ __ ] bored that we're going to watch bots talk to each other for entertainment. I was like, how is human inhuman entertainment not entertaining enough that now you're I do not like that it's becoming popular to just peruse the Reddit of AI **[42:41]** ramblings. They don't know what they're saying, dude. They're just they're again they're autocomplete bots. They're replicating the things that they say. All they're doing, all we're doing as humanity reading them is is watching a mirror. And now I Oh [ __ ] that's actually what TV is, huh? >> Crap. TV is just like a mirror of humanity. Crap. Okay, so it's just a new mirror. Maybe I need to get over it. I I just I'm currently like very I don't like mobile book. I'm like that's the stupidest thing I've seen people use these supercomputers to do is write fantasy madeup autocomplete words to **[43:13]** each other. I'm like, why are you even reading that, man? There's like good stuff to read. Cut that out. Um, but dang, you're right. I guess they could like generally like grow each other into it. That's kind of comes back to like a thing too is like um OpenAI, let's say they they did have this big leap uh or they they had a lead, right? They had ChatGpt 6 before anyone else had it. They're probably working on Chat GPT7. Um, are they not doing the same thing that Mobbook is doing? Are they not creating 10,000 or a million agents in **[43:44]** the cloud that all interface with each other to self-improve so that chatbt6 is just turbo sick? Like why are they not scaling these things against themselves? It seems obvious but somehow we're not seeing the results of that thing. Maybe they are and maybe it's actually not effective. Maybe we'll learn from molt book that eventually remember all those early chat bots that eventually just became like neo-Nazis and then everyone's like turn it off. Turn it off. It went to clickbait. It It went to clickbait to the hole and now it's just like spiraling downwards. Like turn it **[44:16]** off. It's really ugly. >> Is that what they do? Maybe a scale. Is malt book going to be that next week? It's just you're going to see these racist robots. >> It's you know they they always make the joke of like our code is bad cuz we learn from your code. It's like I think everything they do is learning from humanity and there's always going to be dark parts of humanity that are doing [ __ ] they shouldn't be doing and you know dark corners of the internet. So I think yes as soon as they find that they'll go huh this is normal to do. So let's uh let's do that and then they'll **[44:46]** like until you can find a way to get them off the rails and correct them which you can't when they're kind of like doing their own thing. Yeah, it's a problem. Uh, well, I hope your utopia is what happens. And there's just self-improvement so that my requests are fulfilled better every single time. Um, okay. Let's talk about rent a human.ai. >> Yeah, I don't know anything about it. I saw a post the guy made this site to rent humans. So, it's basically like it has an MCP server and you can tell your AI like, "Hey, if you need if you need a human to do something, you can hire **[45:16]** them." And it's really like it's he made it for the memes. and he even says that in the comments like it's not really useful, but it I could see it being a useful thing on occasion of like, hey, I need some like like I need you to go um sign this document in person. Like a real person has to go sign it and I can't do it cuz I'm a robot. So it's like cool. All right, 50 bucks an hour. I'll go do it. Like that kind of thing. I think the >> possibility is awesome. >> Yeah, I think there's there's something there and it would be a rare use case. Like it's not going to be a booming **[45:47]** business, I don't think. But like I think it's an idea that will have a place at some point. >> It's so black mirror. You know, you just wake up and read your emails and you got five bots that need you to do five tasks for the day and so you just do them and just money rolls in and that's just your day. You don't even >> interface with people really. You're just >> Yeah. Um >> Well, I'm personally hoping to uh get my bot set up and have it make some kind of money. I don't know how yet, but um I **[46:18]** mean I think like >> step one, set up the bot. Step two, question mark. Step three, profit, bro. Like >> Yeah. I mean, I think like this this whole idea could could also go dark. Like you could be like, "Hey, what's up? Um I need four humans >> to go rob this guy who I know is going to have this absolutely like hundreds of thousands of dollars on them for whatever reason. And you give me 10% and you guys keep the rest or whatever. and you just do that on repeat. Like, you know, if you give it no guard rails and it doesn't care what the laws are, **[46:49]** it's going to be like, "Yeah, that's clearly I'm going to take every avenue to make money." >> Um, so I think it's almost like that right now as an Uber driver, right? You're just like I you wake up and you check your phone and you've got some orders to fulfill and so then you you drop them off and like your entire interface is not human. It's just you're you're a person doing tasks in the real world for a computer. >> So, we're Yeah. the plausibility of that uh website was very striking. >> Yeah. So, I since you brought up Uber, I want to bring up this post I saw right **[47:20]** before we got on here where they were saying um basically every gig business like Uber and Uber Eats and whatever is going to be replaced by Poly Market. And they were saying the way this works, all right, this is a hypothetical scenario. Okay, >> I don't even know what Poly Market is, so I might need a background. Okay. So, poly market is like you can bet on anything. You can be like it's going to be 70° today or >> and so he's like, "All right, here's imagine this." Uh I'm like, "Hey, I'm **[47:51]** going to bet $15 that four Kiwis will not show up on my doorstep." And it's like, "All right, well, all you got to do is go get four kiwis and right before you drop them off, bet yes and you get $15." So, it's like Uber's dead. You can just do that and just be like anything you want delivered is like done. Like I I think that's a really interesting argument and uh could be plausible as well. And that so I two media comments. **[48:24]** One is there's a South Park episode where um they're betting on Kyle's mom um freaking out about politics and I can't I can't remember exactly. Anyway, but the whole episode has this undertone of people opening their phones and like betting is is she going to do it? She's going to do it. Is she going to do it? She's not going to do it. I got money. She's not going to do it. So like everybody's and then they the kids go to school and everybody's betting on everything. Dude, he's he's totally going to do that thing at lunch. Um anyway, the probability market stuff is pretty wild. **[48:54]** Um, so there's like that aspect of it that's like South Park's kind of already commented on it. You kind of like know it's going somewhere. Um, oh crap. What was the second uh idea was that uh well, I don't know. Must not have been that important. Um, >> we'll never know. >> We'll never know. It definitely uh is a little troublesome though that it's so easy to kind of toss your money at things and to see if they uh happen or not. I don't know. Well, this is how people are making a ton of money with **[49:24]** OpenClaw right now is the poly market stuff. Like, I didn't read fully into it, but I think it amounts to the bot is really fast compared to a human. So, like if there's a thing where the probability is like rapidly shifting the other way, it can like buy the other one before everyone else can or something and then it makes little bits of money. And so, like they do that times millions of bets and you got tons of money. So, >> okay, I remember where I was going with the second point was meme coins. Do you know what a meme like a memecoin is? >> Sure do. Owner of many. **[49:54]** >> Okay. You're owner of many. So yeah, the uh Gas Town person was contacted like, "Hey, you're your Gas Town memecoin has made a bunch of money on this website over here. You're entitled to the money because you created the concept of Gas Town. Sign in, connect a Bitcoin wallet, and we'll dump money into your pocket." >> Nothing suspicious there. Did he >> Nothing suspicious at all. He did it. So yeah, he he did it and then he he got money and it was flowing really his he really well I don't know dude this is **[50:26]** where it gets his [ __ ] internet I don't know you know he just could write whatever he wants on a blog post but anyway you could take your wildly cool trending topic and go monetize it somehow it's kind it's the point of this particular website is like hey if something is becoming that like Ralph Wigum was on there >> other concepts are on there >> um >> I'm forett I know what you're talking about there's like a a thing just for memecoins and it's just for like >> just for memecoins. >> It they'll be popular for like a day and it's Yeah, I forget what >> that's literally what you're betting on. You're betting on popularity. So you're **[50:56]** talking about these probability systems and this one is >> this one is you betting on how. So 67 if you were like early on 67 and you and you bought meme coins of 67 really cheap and all of a sudden 67 super popular. Now you go sell your 67 coin and you made money for 78. What? I don't know. Yeah. You're like, >> so you're like day trading with complete, dude. It's just so unreal. It's But then the thing is just like when Bitcoin came out and you're like, "What is money?" You're like, "Fiat versus currency, bro." And you're like, "Oh crap, that's **[51:27]** actually kind of true. I don't hold any gold. I just have numbers in bank accounts. Oh crap, that's really weird. >> It's all fake." >> We're looking at this the stock market is the same thing. So it's like, what's I'm over here reading about this stuff going, "Oh crap, what's the difference between the stock market and a memecoin? I don't know. >> Not much. >> And I'd really just like >> billionaires propping up one and not the other. That's the difference. >> Yeah. One is indie and one is uh owned by corporate overlords. So yeah, but that's uh but that almost makes it feel like you should go invest in meme coins. But I I don't think that's true. This is **[51:59]** just a gambling market. It's all gambling. All of it is gambling. >> And that drives me nuts. >> Um >> the stock market for sure is like I don't I think I I don't know if I mentioned this to you. I know I've mentioned it on the podcast where I'm like so upset that a stock will be like, "Okay, we uh we expected to grow our revenue by 10%." Right? So, we grew our revenue by 75%. Actually, blew it out of the water. Stock goes down. Like, what? Like, it's like it makes no sense. Like, it's complete luck. There is nothing backing **[52:30]** anything or the insiders have already traded that outcome and it was already inflated. I don't know. But like, oh man, it's fake. >> I got a good segue. >> A good segue to malleable software. So the from the stock market to malleable software. The stock market right now is having crashes in SAS products because investors of software as a service are going uh the future is not software as a service anymore. It's AI building the tool you need on the fly. It's personal software and malleable software. That's **[53:02]** a So SAS products are tanking. Um, and I'm even tempted to sell because I'm like looking at the their reasoning and I'm like, that's actually pretty sound reasoning. I am currently I haven't downloaded a new app months. >> I don't want new apps. I can solve all my problems myself now for like a $5 prompt and query and a little bit of time. Like it's no big deal. So, okay. So if we go from the stock market is crashing in SAS products straight into malleable software which is what PI is uh kind of championing right now is um **[53:34]** self-controlled no dependencies you have these four basic building blocks on this LLM and now you can go do tons of stuff with your open clawbot. Um yeah malleable software I've been thinking about this a lot like it sucks when you had to wait for a dark mode to come out. Malleable software says ask for dark mode and there it is. You just wait. This is done. Um that that feels like the future. So I'm I'm looking at every SAS product right now going I hope you're shaking in your boots. I'm in I'm **[54:05]** a developer and it's writing code. I'm shaking in my boots. You should be shaking your boots. The tax accountant should be shaking in their boots because people are asking Chad GBT to do their taxes now. Like where it's just so overwhelming. So malleable software is interesting. I'm compelled by it, but I also don't like the displacement that's happening. And I so many of these things are hard to balance in uh my mind. >> Well, I think the the software as a service thing just needs to be augmented by AI. Like I just saw a thing on the uh **[54:36]** uh what's it called? JS weekly or whatever the like email newsletter. They were sponsored by some company, I forget, but it was like maybe called Meticulous. Um, and their whole thing is like, hey, we watch every interaction that you do while you're developing your application. Every time somebody goes and clicks something or fills in a form or whatever during development, we write a test for it. >> And so like you never have to write tests or do anything because well there the the where where this would fall down **[55:07]** is for the guy who writes the code, never spins it up, and pushes it to production like which happens more than you would think. um then it would never have a test for it. But like yeah, so um their whole thing is like we're using you know we're observing what you're doing. We're using AI, we're using a mix of um like visual regression testing and some of that and like you know putting all the tools together, putting a on AI on top of it to like augment everything is like like you just need that level up for your SAS product and you need it to **[55:38]** be affordable like more compelling than paying AI to do it all the time for you. Um, you know, just like we talked about AI harnesses, like if you have a really good one that's giving you better outcomes than what you and I get, then it's like that I'll pay you for that. Like if it's helping me be quicker and get better results, um, I don't care what you call it, whether it's a SAS or whatever. I think there's there's a market for that. But yeah, the traditional like we're going to code this for 3 years with a team of hundreds of engineers and we're going to solve **[56:09]** your whatever problem like no, you're never doing that again. Yum. Um, all right. We've got a few more topics and some whatnot. Is there any more whatnot? >> Yeah. Let's Let's just skip to All right. So, we talked about >> No, you got the prediction markets in there. Yeah. >> Yeah. Yeah. What's this down thing? >> Oh, Jim Beam. Okay. So, yeah, this was a contact that I know is telling me that. So yeah, since the tariffs have um cut off a lot of a lot of people are boycotting our booze and um people are **[56:41]** drinking less in general, uh that humongous companies like Jim Beamer are firing people cuz they have surplus of barrels just sitting and they're they're like stopping production. Um, which seems this is like, you know, imagine imagine it's 2026 and you're like an old man and you you've had a business that's been running for 40 years or something like that and you have to look at the market right now and be like, >> I guess we're shutting down like I that **[57:11]** the whole scenario to me just seems very very wild. Yeah. >> Yeah. Um, this is maybe not related to that topic at all, but Caitlin and I have been looking at other vehicle options. Um, we need something electric. And we were talking about the whole thing where like, you know, forever it didn't kind of matter what like internal combustion car you had there. There was enough similarity that like if you were a good mechanic could fix most of the stuff on it, order whatever parts you needed, etc. I feel like with all these **[57:42]** electric cars moving so fast and like you know, the they don't have any of those same parts and whatever. And like like the the reason we we were thinking about this is like we have a Model X right now and they're not going to make those anymore. Like Tesla's not making them. So we're like, well then who's going to fix them? Like are they going to always know >> how it works? Like what about when it's, you know, still working just fine, but the battery is shot and you need a new battery? Like do you replace it or are they like I don't know. I feel like there's a lot like there's going to be a lot of cars that are just junk because **[58:14]** like mechanics can't do it because it's like bespoke to each car company. >> Yeah, that's been one of the threats of I had someone who uh told me never buy a hybrid and they said don't buy a hybrid because now you have two cars to take care of and uh I thought that was a I like this comment from someone. My ex is a model too. That's very good. Very good dad joke and funny joke all in one. I love it. >> Yeah. But yeah, some good comments. I think he was the one I think that uh we posted the uh **[58:45]** >> the short on YouTube of um the use effect is ass thing and he goes, "Hey yo, the uh the audio on this is use effect or something cuz like it wasn't like the mouth wasn't in sync with the audio or whatever." So yeah, shout out to you for the funny comments. >> Oh man, Mr. Ray, you're about to get asked to be on the show. Let's just get your hot takes. It sounds like fun. But yeah, they said don't buy a hybrid because now you have two problems instead of one. And um this this happened like 20 years ago when like BMW, okay, so here small story backstory. I grew up and the first car I **[59:17]** bought was a 1976 Pontiac Firebird um with my own money. Anyway, whatever the thing was you opened up the hood on that car, there was space everywhere. I you could you could easily reach tons of parts. It's like a perfect car to learn how cars work because things were spinning in front of you and you could see their operations, right? It's almost like view source and HTML. It's like right there. Um, very little between you and the uh car. Um, except now so like **[59:47]** BMW started coming out. I remember somebody drove up and maybe it's even like a Honda Accord. Anyway, someone come they popped their hood next to my car and it's just one chunk. There's like covers on it. The whole thing is >> impenetrable. You can't you can't even see inside. And I'm like, >> it's like the uh SpaceX motors. Have you seen that picture where it's like the evolution of the the SpaceX motor where the first one has like all these pipes everywhere and then the middle one has like less pipes and the last one is just like a sleek plastic metal thing. That's what we've been doing with engines is **[60:18]** like because when you pop the hood, you want that satisfaction of like, ooh, that looks clean, but it doesn't look like I could work on it, but it looks clean. >> It looks clean. Yeah, totally agree. They're like, "Hey, they finished the job." Like this. Yeah. My Firebird fell to unfinished. You're like, "I can put my hand in there and get hurt." And they're like, "Yeah, you can. Isn't that cool?" >> That's what you want, though. Like, I had a Mustang and it's like, "Yeah, like this. This is made for you to just take the whole manifold off and put a new one on." Like, dope. >> Yep. Oh, man. That's uh So, I think last week was it we tal Every time I go look **[60:50]** at boats, I think of boats just like websites. Same thing with cars. Uh we started, you know, raw. you could see and understand everything that's going on. But as things evolve and as the complexity grows, we get into a state where you can't manage it anymore. AI right now, our LLMs, they feel small. We're keeping them small. Pi is making small cool again. Guaranteed though, the way that this whole world works is it's going to turn into a state where it's a big black box. It's super powerful. No one knows how it works. We don't know how to hack on it. And now it's causing havoc. And someone turn it off, please. **[61:21]** And we just don't know how to do it. So is that where we're headed? is the same way that cars and houses and all this stuff gets turned into a black box. >> I mean, we know how to handle it. The when if it gets bad enough, most of the AI is hosted here near my house. So, uh, we we'll take it out. >> I would watch that TV episode right now if it was like someone gets in a monster truck. They're like, I'm stealing a monster truck. You're like, why are you stealing a monster truck? He's like, because I'm going to go drive it into the warehouse of all the AI units. You're like, that's kind of badass. I'm **[61:52]** going to watch. Yeah, I I can almost guarantee Ken Wheeler would come with. So, >> so funny him with his backpack full of all his a he's got a vector database in his backpack with AI ready to teach him how to do anything on the fly like he's uh Peter from Heroes. >> Uh so funny. >> Okay. Yeah. So, we are at time. Um is there anything we missed? Anything you want to plug? Oh, uh Cascadia JS. Uh get get your tickets with us. >> Y'all come to Seattle. Yeah. **[62:23]** >> Yeah. We will be MCing. We said last time, but if you didn't hear last time, we will be co-ming. Uh Theo, as you mentioned earlier, will also be there. I saw he's one of the keynote speakers. >> No way. >> I'm excited to high five him and be like, I'm really appreciating your content right now. Thank you. >> Yeah, they've been posting like daily this is a speaker um thing. So, I'm assuming there'll be a this is our MC's about us at some point. But um >> cool. >> I don't know. We'll definitely be there. I booked my hotel and flight. So, we're good to go. >> Awesome. Yep. Stay tuned, y'all. We got **[62:54]** more guests showing up. We have Zach from Warp coming soon. Uh, and other announcements. I can't say anything, but did you get their insiders email inviting you to the special thing that Ben Holmes is in charge of? Maybe he's Maybe he's just going to bring you into that cuz he told me he was going to tell you some stuff. I don't, but I've been talking to Ben Holmes all day today, stinker. I have a meeting with him on Friday cuz I want to be like, tell me what's new and cool and I'm going to learn it on Friday. Dang it. You got an email I didn't get. >> He's not just inviting you to the thing **[63:24]** everyone's going to. >> I don't even know what everybody's going to, man. So, >> I don't either. I got the email like shortly before this, but it's I'm not going to say anything that is said cuz it's it's an early thing. I don't want to spoil their fun, but like >> Dang. Yeah. I don't >> Yeah, they're announcing some stuff and uh they're they're saying like the the people, you know, on the influencer whatever groups like come check it out first and let us know what you think. >> You're in that and I'm not. I think that's why you got it, dude. >> Yeah. despite you referring way more people and doing way more work than I **[63:56]** >> sigh. It's okay world. I'll be patient. >> Yeah. All right. Cool. Well, thanks everyone for listening. If you liked it, please subscribe, come to CascadiaJS, leave us reviews, do all the things, hang out with us. We'll probably also be at Render figuring out details on that. Uh might also be at React Miami TBD. But if you want us at these places, make noise for anyone who is a sponsor and can get us to those places. >> Yep. Definitely make that noise, y'all. >> Yeah. All right. Peace. **[64:29]** You've been watching Whiskey Web and Whatnot recorded in front of a live studio audience. What the [ __ ] are you talking about, Chuck? Enjoyed the show? Subscribe. You know people don't pay attention to these, right? Head to whiskey.fun fun for merch and to join our Discord server. I'm serious. It's like 2% of people have actually clicked these links. And don't forget to leave us a fivestar review and tell your friends about the show. All right, dude. I'm out of here. Still got it.