From Wetware to Software (Presented by Warp)

From Wetware to Software (Presented by Warp) ![rw-book-cover](https://img.youtube.com/vi/d5nKgJz7Kek/maxresdefault.jpg) 🔗 **Source:** [From Wetware to Software (Presented by Warp)](https://www.youtube.com/watch?v=d5nKgJz7Kek) > [!note]+ Document Notes > --- > tags: #ai-development #programming-productivity #claude-anthropic #llm-limitations #developer-experience #agent-orchestration #open-source-alternatives #workflow-optimization > author: Robbie Wagner & Adam Thomas Argyle > source: Whiskey Web and Whatnot > reference: https://www.youtube.com/watch?v=d5nKgJz7Kek > date created: 2025-01-14 02:03 > --- > # 🏷️ Title: From Wetware to Software - AI Development Realities > ## 🧰 Action Inbox > - [H1] Evaluate Claude pricing vs API costs for development teams — financial sustainability #cost-optimization > - [H2] Test memory solutions for long-running agent tasks — reliability improvements #memory-management > - [H3] Research Pretext for typography effects vs accessibility trade-offs — implementation decisions #web-performance > ## 🧭 Table of Contents > - [[#📝 Summary]] > - [[#📖 Main Content]] > - [[#🗒️ Notes]] > - [[#📚 References]] > - [[#🗣️ Paraphrase]] > - [[#📰 Newsletter]] > - [[#✅ Action Items]] > - [[#🧪 Hypotheses]] > - [[#📊 Methodology]] > - [[#❓ Questions]] > - [[#🔮 Future Research]] > - [[#💡 Implications]] > - [[#🧠 Learning Styles Assessment]] > - [[#📊 Accessibility Assessment]] > - [[#🌍 DEIB Assessment]] > - [[#✅ Logical Validity]] > - [[#🛡️ DARVO Analysis]] > - [[#🎯 Conclusion]] > ### 🔗 References > - [Warp](https://warp.dev) > - [Claude Code](https://claude.ai) > - [OpenCode](https://opencode.dev) > - [Pretext](https://pretext.dev) > ## 📝 Summary > - Two experienced developers discuss the current state of AI development tools, particularly Claude and OpenCode alternatives > - Major issues with AI memory persistence and reliability in production workflows > - Anthropic's recent model releases (Mythos) and security research positioning for IPO > - New tools like Pretext enabling advanced typography effects by bypassing browser layout engines > - Increasing productivity expectations from employers despite AI tool limitations > ## 📖 Main Content > This episode covers the evolving landscape of AI development tools from the perspective of working developers. The hosts discuss the Claude source code leak revealing 512,000 lines of code versus OpenCode's 110,000, raising questions about feature bloat versus functionality. > A significant portion focuses on memory and reliability issues with AI agents, particularly OpenClaw's inability to maintain consistent behavior across sessions. Despite repeated instructions, agents fail to remember previous interactions or follow established patterns. > The conversation also covers Anthropic's strategic positioning with Mythos, a new model claiming to find security vulnerabilities rapidly, alongside OpenAI's proposal for AI as public utility infrastructure with 4-day work weeks. > Technical innovations like Pretext demonstrate how developers are bypassing browser limitations through position absolute rendering and off-canvas computation, achieving effects previously considered too expensive for real-time use. > ## 🗒️ Notes > #### ⏱️ [00:00](https://youtu.be/d5nKgJz7Kek?t=0) — [H1] Claude source code leak reveals complexity > - Key idea: Claude has 512K lines vs OpenCode's 110K, questioning feature necessity > - Evidence: "I feel like there's nothing in OpenCode I'm like oh man I wish it would do that" > #### ⏱️ [05:23](https://youtu.be/d5nKgJz7Kek?t=323) — [H2] Anthropic's Apple-like walled garden approach > - Key idea: Banning third-party harnesses to control user experience and data flow > - Evidence: "It feels like a glimpse of how software is going to be built in the future with agents" > #### ⏱️ [12:45](https://youtu.be/d5nKgJz7Kek?t=765) — [H3] Model quality differences affecting productivity > - Key idea: Anthropic models significantly outperform alternatives in coding tasks > - Evidence: "It has been bad like it's back to like using Sonnet or even before that" > #### ⏱️ [18:30](https://youtu.be/d5nKgJz7Kek?t=1110) — [H1] API pricing challenges for power users > - Key idea: Direct API access burns through credits extremely quickly > - Evidence: "That money went out the door really fast and it made my subscription just look really valuable" > #### ⏱️ [25:15](https://youtu.be/d5nKgJz7Kek?t=1515) — [H2] Memory persistence failures in AI agents > - Key idea: OpenClaw consistently fails to remember previous instructions and patterns > - Evidence: "We've done this so many times before. You have to set something to make it a draft" > #### ⏱️ [32:00](https://youtu.be/d5nKgJz7Kek?t=1920) — [H3] LLM fundamental limitations with memory > - Key idea: LLMs are stateless word generation machines, not memory systems > - Evidence: "An LLM is just a word vomit machine. It just does, oh snap, I was given some prompts" > #### ⏱️ [38:45](https://youtu.be/d5nKgJz7Kek?t=2325) — [H1] Anthropic's Mythos security positioning > - Key idea: New model claims rapid vulnerability discovery for IPO positioning > - Evidence: "They spent $50 to find a 27-year old bug in OpenBSD" > #### ⏱️ [42:10](https://youtu.be/d5nKgJz7Kek?t=2530) — [H2] OpenAI's utility infrastructure proposal > - Key idea: Positioning AI as essential utility with 4-day work week implications > - Evidence: "We're going to basically no one needs to work as much anymore" > #### ⏱️ [47:30](https://youtu.be/d5nKgJz7Kek?t=2850) — [H3] Workplace productivity paradox > - Key idea: AI increases output but employers expect more work, not less > - Evidence: "They were like, well, you have AI though, right? So, it's like faster" > #### ⏱️ [52:20](https://youtu.be/d5nKgJz7Kek?t=3140) — [H1] Pretext typography innovation > - Key idea: Bypasses browser layout engine for advanced text effects > - Evidence: "Everything was position absolute and then everything was positioned with transforms" > #### ⏱️ [58:15](https://youtu.be/d5nKgJz7Kek?t=3495) — [H2] Browser performance limitations > - Key idea: Chrome's text rendering constraints versus Safari's unlimited approach > - Evidence: "Safari comes out a couple months later and they go, unlimited lines" > #### ⏱️ [63:40](https://youtu.be/d5nKgJz7Kek?t=3820) — [H3] Agentation visual debugging tool > - Key idea: Chrome extension for visual element selection and prompt generation > - Evidence: "You click where the margin is bad, and you're like, Fix the margin" > ## 📚 References > - [Warp Terminal](https://warp.dev) > - [Oz Agent Orchestration](https://oz.dev) > - [Claude AI](https://claude.ai) > - [OpenCode](https://opencode.dev) > - [Pretext Typography](https://pretext.dev) > ## 🗣️ Paraphrase > - Claude's complexity (512K lines) versus OpenCode's simplicity (110K lines) raises questions about feature necessity versus bloat > - AI memory systems consistently fail because LLMs are fundamentally stateless word generators, not persistent memory systems > - Companies like Anthropic and OpenAI are positioning themselves as critical infrastructure for IPO strategies while increasing productivity expectations > ## 📰 Newsletter > 🚀 **AI Development Reality Check** > This week's deep dive reveals the growing pains of AI development tools. While Claude shows impressive capabilities, its 512K lines of code versus OpenCode's 110K raises important questions about tool complexity. > **Key Takeaway**: Memory persistence remains AI's biggest challenge. Despite repeated instructions, tools like OpenClaw fail to maintain context across sessions, forcing developers to constantly re-teach basic patterns. > **What's Working**: Direct subscriptions to Claude and Anthropic models continue to outperform alternatives, though API costs can quickly spiral for power users. > **Innovation Spotlight**: Pretext demonstrates how creative developers bypass browser limitations, achieving real-time typography effects through position absolute rendering. > Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) > ## ✅ Action Items > #### DevRel > - Document AI tool reliability patterns: create guides for handling memory failures > - Workshop planning: "AI Development Workflows" comparing subscription vs API approaches > - Community feedback: survey developers on AI productivity expectations versus reality > #### OSPO > - License scanning: review AI-generated code provenance and attribution requirements > - Security protocols: establish guidelines for AI vulnerability discovery tools like Mythos > - Tool evaluation: assess open source alternatives to proprietary AI development platforms > #### WordPress > - AI integration: explore Pretext typography capabilities for WordPress themes > - Performance budgets: measure AI-enhanced development velocity against output quality > - Documentation: create patterns for AI-assisted WordPress development workflows > ## 🧪 Hypotheses > - AI subscription models provide better value than API access for consistent development work > - Memory persistence issues stem from fundamental LLM architecture rather than implementation bugs > - Browser performance constraints drive innovation in alternative rendering approaches > - Productivity expectations increase proportionally with AI tool adoption regardless of actual efficiency gains > ### 📊 Methodology > - Cost analysis comparing Claude subscription versus API usage over 30-day periods > - Memory reliability testing across different AI platforms using standardized task sequences > - Performance benchmarking of Pretext versus native browser text rendering > - Developer productivity surveys before and after AI tool adoption > ### ❓ Questions > - How will memory persistence be solved without fundamental changes to LLM architecture? > - What happens when AI vulnerability discovery tools become widely accessible to bad actors? > - Will the productivity gains from AI development tools ever translate to reduced working hours? > ### 🔮 Future Research > - Memory augmentation systems combining LLMs with traditional database approaches > - Browser APIs specifically designed for AI-enhanced development workflows > - Economic models for AI-assisted work that don't simply increase output expectations > ### 💡 Implications > - Practice: Developers need fallback strategies for AI memory failures and cost management for API access > - Policy: Organizations must establish guidelines for AI tool usage and productivity expectations > - Security: Vulnerability discovery acceleration requires coordinated disclosure frameworks > - Education: Training programs need to address AI tool limitations alongside capabilities > ### 🧠 Learning Styles Assessment > Reading level estimate: Grade 12-14 (technical/professional) > #### 👀 Visual learners > - Present: Code examples, tool interface discussions > - Improve: Screenshots of Claude/OpenCode interfaces, workflow diagrams > #### 🔊 Aural learners > - Present: Podcast format with detailed explanations > - Improve: Clearer audio transitions between topics > #### 🗣️ Verbal learners > - Present: Technical terminology with context, analogies to traditional development > - Improve: Glossary of AI development terms, more explicit definitions > #### 🧩 Kinesthetic learners > - Present: Hands-on tool usage descriptions > - Improve: Step-by-step setup guides, reproducible examples > #### 🧠 Logical learners > - Present: Cost comparisons, performance metrics, systematic analysis > - Improve: Decision trees for tool selection, quantified productivity metrics > #### 🧑‍🤝‍🧑 Social learners > - Present: Community discussions, shared experiences > - Improve: Developer experience sharing prompts, collaboration patterns > #### 🧘‍♀️ Solitary learners > - Present: Individual workflow optimization > - Improve: Self-assessment tools for AI development readiness > #### 🌿 Naturalistic learners > - Present: Evolution of development tools, ecosystem thinking > - Improve: Historical context, development tool lifecycle analysis > ### 📊 Accessibility Assessment > #### Screen reader semantics > - Present: Structured content with clear headings > - Improve: More descriptive link text, better landmark usage > #### Keyboard navigation > - Present: Standard podcast controls available > - Improve: Transcript search functionality, chapter navigation > #### Contrast and readability > - Present: Standard text formatting > - Improve: Code syntax highlighting for accessibility > #### Text resizing and reflow > - Present: Responsive transcript layout > - Improve: Custom text sizing controls > #### Media and captions > - Present: Full transcript available > - Improve: Timestamped captions, downloadable transcript > #### Forms and interactions > - Present: Basic playback controls > - Improve: Accessible speed controls, chapter selection > ### 🌍 DEIB Assessment > #### Cultural breadth > - Present: US-centric development perspective > - Improve: Global developer experiences, international tool availability > #### Inclusive language > - Present: Technical discussion without exclusionary terms > - Improve: Avoid assumption about team structures, development contexts > #### Neurodiversity support > - Present: Structured conversation with clear topics > - Improve: More explicit topic transitions, content warnings for rapid topic changes > #### Representation > - Present: Two experienced developers sharing perspectives > - Improve: Broader range of developer experience levels, different technical backgrounds > #### Psychological safety > - Present: Honest discussion of tool limitations > - Improve: More explicit acknowledgment of learning curves, skill level variations > ### ✅ Logical Validity > #### Strong reasoning > - Deductive strength: Clear cause-effect relationships between tool complexity and usability > - Inductive strength: Pattern recognition from multiple AI tool experiences > - Causal reasoning: Well-supported connections between pricing models and usage patterns > - Analogical strength: Effective comparisons to historical browser performance issues > #### Fallacies > - Availability heuristic: Recent negative experiences with OpenClaw may overshadow overall utility > - False dilemma: Presenting subscription vs API as binary choice when hybrid approaches exist > #### Tone of voice > - Overall tone: Experienced, candid, slightly frustrated but constructive > ### 🛡️ DARVO Analysis > - Deny: No denial patterns observed > - Attack: No attacking of accusers present > - Reverse victim/offender: No victim/offender reversals noted > - Emotional cues and rhetoric: Frustration expressed toward tools, not people; constructive criticism maintained > ## 🎯 Conclusion > AI development tools are in a rapid evolution phase where capabilities are impressive but reliability remains inconsistent. Memory persistence represents the biggest unsolved challenge, forcing developers to work around fundamental LLM limitations rather than rely on them. > The productivity gains are real but come with increased expectations from employers, creating a productivity treadmill effect. Tools like Pretext demonstrate how innovation often requires bypassing existing systems entirely. > Developers should focus on hybrid approaches using multiple tools, direct subscriptions over API access for consistent work, and building workflows that account for AI limitations rather than assuming perfect reliability. > ## 🧱 Refactor Appendix > ### 📚 Literature Split Pad > ### [H1] AI memory systems require architectural solutions beyond LLMs > > [!quote] From Wetware to Software by Robbie Wagner & Adam Thomas Argyle via Whiskey Web and Whatnot. Published 2024. Accessed 2025-01-14. > > "An LLM is just a word vomit machine. It just does, oh snap, I was given some prompts. I need to make another word and another word and another word." > Core limitation of current AI development tools requiring external memory systems. > [view highlight](https://youtu.be/d5nKgJz7Kek?t=1920) #xref #articleseed ID:rw_ai-memory-limits > ### [H2] Browser performance constraints drive alternative rendering innovation > > [!quote] From Wetware to Software > > "Everything was position absolute and then everything was positioned with transforms. You're entirely on the GPU." > Pretext bypasses browser layout for performance, similar to Famous.js approach. > [view highlight](https://youtu.be/d5nKgJz7Kek?t=3140) #xref ID:rw_browser-bypass > --- > ### 🔖 Atomic Split Pad (Zettel Candidates) > #### [Z1] AI productivity gains increase workplace expectations > - Applies to knowledge work across industries using AI tools > - Evidence from multiple developer experiences at 18:30-47:30 > - Links: [[Productivity-Paradox]] [[AI-Workplace-Impact]] > #### [Z2] Third-party AI harnesses face platform restrictions > - Anthropic's ban affects OpenCode, PI, and similar tools > - Business model protection strategy at 5:23-12:45 > --- > ## 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 [00:09] Party. [00:12] >> 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. [00:23] >> That's right, Charles. We drink whiskey [00:25] and talk about web development. I mean, [00:27] it's all in the name. It's not that [00:29] deep. [00:30] >> This is Whiskey Web and Whatnot. Do not [00:33] adjust your set. [00:36] Whiskey Web and Whatnot is brought to [00:38] you by [00:42] This episode is brought to you by warp [00:43] atworp.dev. Warp is now open source and [00:46] warp is built using Oz publicly for us [00:49] all to see. And it feels like a glimpse [00:51] of how software is going to be built in [00:52] the future with agents except that's [00:54] also now. But what's Oz? the cloud agent [00:57] orchestration platform by the warp crew. [00:59] You got to check it out. Warp has also [01:00] made their cloud integration even [01:02] better. Find Ben Holmes and ask about [01:04] the rad codeex integration they just [01:05] added too. You can also publish an agent [01:08] session from claude codecs or open code [01:10] to the cloud with one click and then [01:12] monitor or steer that agent from your [01:14] phone browser or some other computer. It [01:16] is sick. All right, go ahead and close [01:18] CMX because warp tabs are now on the [01:20] left side. They look like projects and [01:21] each pane is visible within that tab [01:23] project group. I've got four tabs open [01:25] right now. Each tab has anywhere from [01:27] one to four terminals in it, keeping [01:28] sessions organized. And that visibility [01:30] into the PES without having the tab open [01:32] is awesome. Warp updated supported [01:34] models, too, adding Kimmy K2.6, GLM 5.1, [01:37] Miniax, and Coen. Check out warp at [01:39] warp.dev or oz.dev. Sweet. Back to that [01:43] episode. Hey, what's up everybody? [01:45] Welcome to Whiskey Web and [01:48] whatnot with your hosts Robbie the [01:50] Wagner and Adam Thomas Argyle the nerd. [01:53] You are absolutely right. [01:57] >> It'll get old one of these days. It's [01:58] not there yet. [01:59] >> Yeah. How you doing, Adam? [02:01] >> Oh, man. Good. You know, classic. Uh I'm [02:04] more empowered and faster than ever [02:06] before and more uh anxiety. I'm working [02:08] harder than I ever have before. So, the [02:10] promise is completely unfulfilled and [02:12] I'm waiting for Alanis Morset to sing a [02:14] song about it. [02:17] >> Hey, well, you can use Gemini to make [02:19] that song. [02:20] >> Sure can. Oh, snap. That would be really [02:22] funny. I had so much fun trolling my [02:24] friends with that. Like cuz I got [02:25] friends that are like playing uh Slay [02:27] the Spire during video calls. And so I'm [02:29] like, "Hey Gemini, write a rap about [02:30] this fool who thinks he can be in a [02:31] video call and play Slay the Spire at [02:33] the same time." And it wrote it and it [02:35] had these like great uh lines in it. And [02:38] oh man, it's been so fun. [02:40] >> Nice. Nice. Well, uh yeah. So before we [02:44] pour this whiskey, do you want to tell [02:46] the folks what we might be talking about [02:48] today? In today's episode, we're [02:50] covering the Claude leak, open claw [02:52] replacements, going from wetwware to [02:54] software, pretext, inpage agents, [02:56] instant translation, killing entire SAS [02:58] companies, AI music, second brains, [03:00] ultimate hacker death stars, celebrity [03:02] deathmatch, agent battles, personified [03:04] agents over video calls, and are LLM's [03:07] AGI yet? Come on, we're ready. Or are [03:09] we? [03:12] Yeah, a lot to cover. So, we better [03:14] grease the wheels with some Artbeg. Wee [03:18] beasty. [03:18] >> Wee Beasty. So fun. I can just beasty. [03:21] >> It's a wee. [03:22] >> More Italian then. [03:25] >> Turn into Luigi over here. [03:27] >> Yeah. [03:30] >> Oh, I'm ready for a scotch in my life [03:32] again. Let's do it. [03:34] Oh, speaking of scotches, I got the uh [03:37] my wife is doing a Costco order and I [03:39] was like, "Hey, I I had the the Kirkland [03:42] brand scotch. a week ago. It was pretty [03:44] good. Will you um get me one? And she's [03:45] like, "Yeah, there's like the regular [03:47] scotch or the 12-year Kirkland brand." [03:48] I'm like, "Yeah, go to the 12 12 year." [03:50] It was $35. And I was like, "That's so [03:53] cheap. Get get it." It showed up. It's [03:55] absolutely massive. The thing is like [03:58] this, it's a Costco sized huge cir. I'm [04:02] like, "Oh my goodness." Like, it doesn't [04:03] even fit in my cupboard. [04:06] >> Oh yeah. That's the way their vodka is, [04:08] too. It's like [04:10] >> um you know, it's like, you know, it's [04:12] like way too big. [04:13] >> So big. I'm like, it's like a 2 L bottle [04:15] of scotch. And I'm like, this is so [04:17] American to just have that much uh [04:19] scotch delivered to my door. It's just [04:21] hilarious. [04:23] >> Yeah, dude. [04:23] >> This smells good. [04:25] >> Yeah. So, reminder for our loyal [04:27] listeners, we now talk about the whiskey [04:29] at the end. So, if you're interested in [04:31] that, listen till the end. And if you're [04:34] not interested in that, stop listening [04:37] when we start talking about it. [04:41] All right. Yeah. So, one of the um news [04:46] things that is a little bit old now. [04:48] It'll be really old by the time you hear [04:50] this. I'm going to start releasing [04:51] episodes a lot more often hopefully so [04:54] that we can't can stop saying that. But [04:56] Claude Code source was leaked which uh [05:00] is hilarious. I saw a fun tweet about [05:03] like um you know they want you to pay [05:06] more for them to like review your PRs [05:09] also and that the fact that they could [05:11] like ship this out and not notice is [05:14] like just hilarious. [05:16] It is funny. I love all the things [05:18] people went looking for. Um I was like [05:20] oh the source is leaked and you know I [05:22] didn't get any creative ideas and West [05:24] Boss is like I wanted to know how many [05:26] of the words it says when it's thinking. [05:29] Yeah, [05:29] >> I'm like that's funny. [05:31] >> That was a good one. [05:32] >> Yeah, I saw that. I think it had like [05:34] 512,000 [05:36] lines of code or something. And I think [05:39] uh Open Code has like 110,000. These are [05:42] um I'm not sure the exact numbers, but [05:44] Dax had posted about it and they were [05:46] like, "Oh, I'm getting loc [05:51] to up our our lines of code." And um so [05:54] yeah, I think that's interesting that [05:56] they have so much in there. I feel like [05:57] they have a ton of stuff that people [05:59] don't need. Like I think this goes back [06:01] to the thing STAX is saying about like [06:03] don't just ship stuff cuz you can like [06:05] make sure it's something people want to [06:07] use cuz I feel like there's nothing in [06:08] open code or that's not an open code. [06:11] I'm like oh man I wish it would do that. [06:12] I don't know. How do you feel? [06:15] Less is more. Although you know like I [06:19] stopped using Pi as much recently and [06:21] went back to Claude. So, it's like I I [06:23] loved the idea of, you know, training a [06:25] puppy and growing this thing with me and [06:27] starting small. And eventually, I was [06:29] just like, I don't know, some of the [06:31] noise that that caused, I was like, no, [06:33] put the batteries in. Batteries [06:35] included. I don't want to think about [06:36] it. Just work cuz I want to sit down and [06:38] I just want to get results. And so, um, [06:41] I'm going to waffle on this forever just [06:43] as we always do, right? This pendulum [06:45] will swing. Um, but yeah, if it's got [06:48] that much code in it and it's handling a [06:50] lot of my cases, then it's all good. I [06:52] do like their team implementation. Um, [06:54] and I like a lot of the, you know, they [06:56] they kind of keep hardening some of the [06:58] better patterns in the things. So maybe [06:59] that's why it's growing. So instead of [07:00] relying on plugins, uh, they just build [07:02] it in. It's a very Apple mentality. I'm [07:04] feeling very much like Anthropic is [07:06] turning an Apple uh, page here. They're [07:09] like learning from them like, "Oh, we [07:10] need to create the walled garden. We [07:11] don't want people to be reaching outside [07:13] of our system, which is also why they [07:15] closed out OpenCloud." We'll get to [07:17] that. But I'm just getting a lot of [07:18] Apple and maybe that's cuz they want to [07:20] IPO and they're like, "We need to we [07:21] need to trap people." Uh, which I hope [07:24] they're not saying that, but they might [07:25] be. [07:26] >> So, yeah. I'm I don't want to say [07:29] anything super bad cuz, you know, I do [07:31] have their free max plan. I'm enjoying [07:34] it. Thanks. Thanks again for that. Um, [07:38] however, [07:40] it's it doesn't feel great to use. just [07:42] like, yes, it has those fun words of [07:44] like whatever it's doing, uh, [07:46] discombobulating or whatever, like every [07:49] time. Um, but I feel like I'm [07:52] >> I'm not clear on exactly what it's doing [07:54] all the time. And maybe that's by [07:57] design. Um, but I feel like in open [07:59] code, I know for sure if I'm planning or [08:01] not. I know like um what it's thinking [08:05] about. I can like I feel more in [08:07] control. And maybe it's just because I [08:09] haven't used claude code as much, but um [08:12] one thing is for sure, you need to be [08:14] using the anthropic models. They are so [08:17] much better than all the other ones. Um [08:20] so we lost the ability to use them at [08:22] work and I've been having to use uh GPT [08:25] 5 5.4 or codeex 5.3, which I don't know [08:29] why they're on different numbers. Um [08:31] whatever. Um anyway, and it has been bad [08:38] like it's it's back to like using Sonnet [08:40] or even before that like like you know [08:43] back when we used to poo poo on AI. I [08:46] feel like that's what the results I'm [08:47] getting now of like you you know nothing [08:50] like I'm going to hurl obscinities in [08:52] your direction because I'm so tired of [08:55] you not doing what I asked and like it [08:58] it also oversteps or something. It's [09:00] like it does too much. And maybe that's [09:02] part of I need to give it more [09:04] instruction. But like I think by default [09:08] there was less getting done and it was [09:10] like more powerful correct code getting [09:12] done with Opus then like this just tries [09:15] to vomit like more code to solve the [09:17] problem and that's not how it should [09:19] work. Yep. I use alternative models to [09:23] review the output of Opus, but Opus is [09:26] always the reliable uh good one. And [09:28] that's why I continue to pay their their [09:31] subscription model. I like those models. [09:32] Those are the ones I want to have the [09:34] most access to. Um and that's why I was [09:38] so gut-wrenching when they said I [09:39] couldn't use PI to access uh those [09:42] models anymore. Can you from open code? [09:43] You have to use open router. How are you [09:45] using Opus from Open Code? [09:48] Uh, I have open code Zen or black. Open [09:52] code black, sorry. Um, whatever it is, [09:56] different words. Um, and that still lets [10:00] you use it. It's, um, you know, similar [10:01] to paying for your cloud subscription. [10:04] You can pay like 100 bucks a month, 200 [10:06] bucks a month, like whatever tier you [10:08] want. Um, I don't know how the limits [10:11] compare. I would imagine the limits are [10:13] much lower cuz I think they have to pay [10:15] API prices because they got shut down [10:17] from like [10:18] >> that [ __ ] goes fast, man. I tried it. I [10:20] um and that money went out the door [10:23] really fast and it made my subscription [10:24] just look really valuable. [10:26] >> Um [10:27] >> Mhm. [10:28] >> because they banned third party [10:29] harnesses. Yeah. [10:31] >> Yeah. Yeah, if I'm using it all day, [10:33] like I I'm going to sit down and I'm [10:34] going to use open code with um with some [10:37] anthropic models and I'm going to code [10:39] something today, [10:41] maybe a few somewhere between a few [10:43] hours in to like definitely before the [10:45] end of the day, it's like you're at your [10:47] limit and like maybe not even just like [10:48] the daily or what like the the weekly [10:50] limit or whatever and I'm just like oh [10:52] [ __ ] like so we're just stuck. I'm just [10:55] not going to do anything else. So then [10:56] you know luckily I have many other [10:58] options. I use warp someum. I use um [11:01] Claude code directly like I just bounce [11:03] around as until I hit limits. Um that's [11:05] how I've been doing it. [11:07] >> Yep. I do that with Warp and Claude. Um [11:09] and Warp I'll hit my credit limits [11:11] pretty quick because I'm only paying 20 [11:12] bucks a month or so. Um and I still like [11:15] Warp in a whole bunch of ways. Um so I'm [11:17] going to keep paying that one. Um but my [11:19] Claude one, I'm only at the $100 a month [11:21] one and I haven't reached those limits [11:24] one time. Um I wouldn't say I'm like a [11:26] power user on my personal time. Um, but [11:29] I definitely I'll leave it running and [11:31] it'll do things for 45 minutes and it [11:33] won't even, you know, get past 25% of [11:35] what my limit is. So, I'm feeling pretty [11:37] good about um staying inside of that [11:40] ecosystem for now. Um, but it did make [11:43] it hard with open clause. They said no [11:44] pi. Yeah, no open code, no third party [11:46] harnesses at all. You better figure out [11:48] uh how to use it in some other way. And [11:50] I yeah, I tried the keys and just ate up [11:53] 10 bucks in like a couple queries and I [11:55] was like, well, this is just not going [11:56] to work. Um, [11:59] >> but I did solve it for OpenClaw and got [12:02] it worked out. So, they there's a I and [12:04] here's the thing. I don't know if it's a [12:05] loophole or not yet. The the way that [12:08] they're saying this is no third party [12:10] harnesses, which is open code, uh, PI, [12:14] uh, codeex, I suppose, and other things [12:16] that or even co-pilot if it's the [12:17] harness. You can't use their models [12:20] through some other window. This is, [12:22] that's what I mean. It's so Apple. It's [12:23] like Safari. It's like, "Hey, our uh [12:25] model is Safari and you can only use it [12:27] on an Apple product or I guess it's [12:28] iMessage. You can only iMessage from an [12:30] iPhone type of deal." [12:31] >> So, it seems weird though, right? [12:33] Because like, sorry to derail you, but [12:35] like [12:35] >> Yeah. [12:36] >> Um, essentially a harness is just [12:39] feeding more data into the prompt, [12:42] right? So, like [12:44] um how is that different than me [12:46] defining my own like agents MD or [12:50] whatever like my own instructions? It's [12:52] like what if I just go cool I'm going to [12:53] take exactly what open code has been [12:55] telling you and put it all in here. [12:58] That's fine. You're just you just don't [12:59] want me to have a different UI where it [13:02] looks a little different. They want some [13:05] control. They want to know what what was [13:07] that went in there before it turned into [13:09] the actual query. Um and that's kind of [13:12] why it's conf, you know, confusing about [13:14] what I have set up is it's using the [13:16] cloud CLI. So they're, you know, they [13:19] allow you to programmatically access [13:20] models through the cloud CLI, which I [13:23] guess is just not a harness. Or if [13:25] you're, if you put logic in front of the [13:27] CLI call, it's not a harness. So we need [13:30] to really dig in at some point and [13:32] figure out what is a harness uh and how [13:34] is that different from just a proxy [13:37] because it feels like a proxy and like [13:38] so you have open router. Open router is [13:40] a proxy. You have all these things that [13:41] are sort of like decorating the request [13:44] before it goes in there. But how is a [13:45] harness? And I think what it was is some [13:47] harnesses really put a lot in front. Um, [13:51] and that was what their big complaint [13:52] was with open claws. It was just packing [13:54] in because it would heartbeat and it [13:56] would do and every heartbeat had all [13:58] this context. And so they were just [13:59] getting these like small queries that [14:01] had huge amounts of context in there and [14:03] they were like, "This is just not cool." [14:05] And maybe when they owned the harness, [14:07] uh, they can mitigate a lot of that [14:08] thing. I'm not sure. Um, yeah. Nice. Mr. [14:12] Ray's like, "Hey, we sound so slow." I [14:14] do the same thing. I meet people in real [14:15] life and I'm like, "Whoa, I listen to [14:16] you on the podcast and you're like [14:19] snails right now. I can't even believe [14:20] it." Um, but yeah, I found Claude uh [14:22] Claude Claw, which acts a lot like So, I [14:25] still interface through Discord. I still [14:27] send all my messages. I can still um [14:29] thread out and um you know, get all my [14:31] sub agents and everything. Uh except I'm [14:34] just not going through PI anymore. I'm [14:36] pinging a um bun server and the bun [14:40] server then forwards it to the CLI the [14:43] cloud CLI and then it gets a response [14:45] and pipes it back into Discord. So it's [14:46] some it's so similar that I'm like this [14:49] is sanctioned but the other one wasn't. [14:51] What's the difference? Is this one going [14:52] to get canceled at some point? I'm also [14:54] trying Claude Co-work because it makes a [14:56] lot of promises like OpenClaw and that [14:59] has not been as successful so far. It's [15:02] anyway that's quite a predicament we're [15:04] in. [15:05] Yeah, I mean [15:07] I think they should have just put out [15:09] their own open claw basically. [15:13] Like I don't I mean maybe they're [15:15] working on it and like there wasn't an [15:17] exact um need for that sort of product, [15:22] but you know they they should have just [15:25] acquired it or built their own thing. A [15:28] lot of people are building the consumer [15:29] version of this now that we were talking [15:31] about earlier like [15:33] um uh sentient something and uh I don't [15:39] have the notes up. Hold on. Hark. Yeah. [15:42] Sentience or Yeah. Sentience and [15:45] >> Yeah. Cension stock company and Hark. [15:47] Yeah. These are both promising to be [15:49] consumer ready uh open class for you. [15:52] Yep. [15:53] >> Yeah. So, I think like this is going to [15:54] be the next thing that people are going [15:56] to try to charge you tons of money for [15:58] is how do we get that magic of OpenClaw [16:01] and one make it easy. There's no setup. [16:04] You just use it. Two, uh, make it [16:07] actually remember and work, which [16:08] openclaw never ever ever one time has [16:12] done for me. And I got so heated with it [16:15] the other day. Um, like I was like, [16:18] "Hey, this thing that we do every week, [16:20] my dude, let's let's make some uh [16:22] episode art." It's like, "Cool, cool. [16:25] Um, all right, here it is." I was like, [16:26] "All right, cool. That looks like [ __ ] [16:28] Why did you do it that way?" And it's [16:30] like, "Well, uh, I don't know. Let me [16:32] like," it didn't ask me any clarifying [16:34] questions and just started trying like, [16:36] "Let me try uh these like 15 different [16:39] image generation things that none of [16:41] work cuz I don't have any API keys for [16:42] or anything." And um it was like uh well [16:45] yeah I couldn't get any of that. So what [16:47] if I just uh give you some text to put [16:49] in somewhere and try to prompt it to [16:50] give it to you and like whatever. And so [16:53] that didn't work. And also another thing [16:55] it does all the time we post a post [16:58] bridge and like whatever. And I'm like, [16:59] "All right, every time you post a post [17:01] bridge, like if there's one thing you [17:03] need to remember, like put it in every [17:06] memory spot you're going to check, make [17:09] it a draft because you might [ __ ] it up [17:12] and I want to review it and say, "No, [17:14] no, that sounds stupid." [17:15] >> I saw this post it made. [17:17] >> Yeah. [17:17] >> Yeah. It made like, okay, so it posted [17:20] it goes, "Cool. I posted everywhere." [17:22] And I was like, "Was it a draft?" I was [17:23] like, "Oh, dang. It wasn't a draft. Let [17:25] me like go edit that one." I couldn't [17:27] edit it, so I posted another one. It [17:28] wasn't a draft either. I was like, "Damn [17:31] it, now we got two out there." And then [17:32] I was like, "All right, can you make it [17:34] a draft?" And it goes, "Uh, actually, [17:35] the API doesn't support making it a [17:37] draft. There's no way to do it." I was [17:39] like, "We've done this so many times [17:41] before. You have to set something to [17:43] make it a draft." It's like, "No, it [17:45] doesn't support it, man." And I was [17:47] like, "Okay, there's an open claw skill [17:50] for this." Like in the what do they call [17:51] it? The open claw [17:53] >> registry or whatever. [17:54] >> Clawub or whatever. [17:55] >> Yeah. Claw. [17:56] >> Yeah. So I was like, "You've installed [17:58] this skill. Go read it." And he goes, [18:00] "Oh [ __ ] yeah, that does tell me all [18:02] the stuff I can do." And I was like, [18:03] "Yeah, okay. Can you do it?" It's like, [18:04] "No." And I was like, "What in the [18:06] world?" Like, and so I go, "Set is_draft [18:10] to true in your uh post." And it goes, [18:14] "It works." And I was like, "Yes, you [18:16] [ __ ] dumbass." Like, how are you [18:19] having this much trouble? We've done [18:20] this so many times and I had to manually [18:22] tell you exactly what to do because you [18:24] couldn't figure it out. And like, man, [18:27] the magic is wearing off on this thing. [18:29] Like, I know you've been having a lot of [18:30] open claw success, but [18:32] >> it's been getting worse and worse for [18:33] me. [18:34] >> I've had I've had memory issues, plenty [18:35] of them. And that's why you see there's [18:37] a lot of people competing in the memory [18:38] space right now. They There was even a [18:40] celebrity that came out with a aentic [18:42] memory system recently. [18:43] >> Did you see that one? And I was like, [18:45] "Wow, we're at the point now where like [18:47] celebrities are making [ __ ] for AI. [18:49] That's pretty hilarious." Um, but yeah, [18:51] and I' and I've been working on a second [18:52] brain, and the second brain is still [18:54] just as annoying. It's all markdown [18:55] files that live on the thing. Um, and or [18:59] on the thing. It's like my machine right [19:00] here. I'm like pointing to it. But [19:01] anyway, it's like I'm like, "Hey, hey, [19:03] I'm going to the baseball game. Uh, tell [19:05] me all the kids families that I'm going [19:06] to see there. I need to know the [19:07] parents' names." And it's like, "Uh, [19:10] dude, it lists out baseball team [19:11] players." It goes, "Yeah, here's some [19:13] baseball players." And I'm like, "What [19:15] are you talking?" No, I specific, you [19:18] know what else does this to me, too, is [19:19] like Google Home. We have a Google Home. [19:21] And I'm like, "Hey, play my playlist [19:23] called Morning Kuners." And it goes, [19:25] "Playing the playlist something [19:27] something kuners." And I'm like, "No, [19:28] no, no, no, no. Stop my playlist. I [19:31] don't give a [ __ ] about everybody else's [19:32] playlist. Just just mine. So, would you [19:34] play my, you know, query first for mine, [19:38] then match the words, play it very [19:40] easy." Uh, and still won't do it. I'm [19:42] just like, "Fine, I'll whip out my [19:43] phone." Geez. Um, but yeah, people are [19:45] trying to figure out this memory thing [19:46] because it's really frustrating. These [19:48] things are innately not memory. They're [19:50] not supposed to have memories. An LLM is [19:52] just a word vomit machine. It just does, [19:54] oh snap, I was given some prompts. I [19:55] need to make another word and another [19:56] word and another word. And we're trying [19:58] to give it a sense of this by [20:00] frontloading it. And it's just not [20:02] working. Um, and [20:06] >> it should be part of the initial [20:08] harness, right, of like if you're [20:09] openclaw, that's supposed to be what [20:11] like or I guess pi under the hood, the [20:13] learning thing, right? [20:14] >> Yeah. is like that should say, okay, [20:18] we're all computer programs at the end [20:20] of the day, right? So, when you start [20:21] the program, look at the [ __ ] memory [20:25] and the like just research what you're [20:27] supposed to do and make sure that you [20:29] remember what we talked about you [20:31] remembering before you ever try to do [20:33] anything else. That should be easy to [20:35] program, I would think. Why when I go [20:38] remember this, remember this, remember [20:39] this. Is it like, [20:41] no, no. I mean, someone was tweeting me [20:43] that on Blue Sky. They're like, "Why did [20:45] you expect a non-deterministic system to [20:47] be somehow deterministic?" And I'm like, [20:50] "Well, because it's kind of [20:51] deterministic in a lot of scenarios. I [20:53] don't think it's unrealistic for me to [20:54] expect it to do these things. I get what [20:56] you're saying, but also there's enough [20:58] guard rails that we can put this thing [20:59] on rails. It's on rails already. All I [21:02] need to do is make sure I give it enough [21:04] descriptions that it it stays on rails." [21:06] But [21:07] >> yeah, it just mixed like like there [21:10] needs to be more [21:12] explicit instruction [21:14] run a script kind of things before the [21:18] AI oh yeah llm like but I guess it's [21:22] trying to understand what I asked for in [21:23] the first place before it can load the [21:26] relevant information. So like if you [21:28] made a like slash command it' probably [21:29] be better than if I'm like make episode [21:32] art then it has to go oh let me like [21:34] search for how to make episode art like [21:36] you know what I mean? So maybe that's my [21:37] fault of not being explicit with my [21:39] commands. [21:40] >> I don't know. I think it's also like [21:41] this Claude Claw system that I ended up [21:43] setting up. Um, it's programmatic, [21:45] right? It's calling the CLI. So, it has [21:47] it has an opportunity to take in what I [21:49] said, break it into pieces, do something [21:51] more programmatic with it before it [21:53] eventually hits the cloud CLI or the [21:55] Yeah. Um, and it does a little bit [21:58] except you know what's back is the [22:00] posting to the general channel. And I'm [22:03] like, no. I'm like, no, this thing is [22:05] just going to haunt me forever. And so, [22:06] like I'm even in there, I'm looking at [22:08] the cloud MD file. that says like never [22:10] post to the general basically I'm like [22:12] if you are replying [22:14] never ever the general channel you're [22:16] replying to something there there's [22:18] something that you are innately [22:21] messaging back to never use the general [22:23] channel for that and then you know what [22:25] it started it always gives me different [22:27] solutions to it's like I put one in the [22:29] reply and one in general and I'm like [22:31] well that's stupid I just needed the one [22:33] in the thing [22:34] >> so yeah I don't know what these memories [22:36] >> does that [22:37] >> yeah clause does that the art will be in [22:40] the general channel, but like the the [22:43] talking of like it'll be like I [22:44] generated this stuff in the thread, but [22:46] then it'll post the like art in the [22:47] general channel. So, it's like it's [22:49] almost there. But the threads are super [22:51] important actually because I have the my [22:53] open claw at home, which I don't use for [22:55] anything um because it cost me so much [22:58] money and was so shitty. But um I think [23:01] it was because I was using iMessage and [23:03] I think in iMessage with no threads. [23:06] >> It's reading the entire iMessage history [23:09] which included me asking like could you [23:11] look up the uh you know API shape of a [23:13] couple things which is like 20,000 [23:15] tokens per one. And so every time you [23:17] did anything it was like you're over [23:19] your tokens as soon as you asked it like [23:21] one question. Um [23:22] >> yCL did that to me. [23:24] >> Yeah. When I first set it up. So, I [23:26] think the the Discord thread thing is is [23:28] a must or something that threaded to [23:30] where it's not going to read an entire [23:32] channel to like understand what you're [23:34] trying to do. [23:35] >> Yep. So, we got our wetwware. We're [23:36] trying to turn it into software and I'm [23:38] just glad I'm not like trying to build [23:40] one of these products cuz look at us. [23:42] We're just consumers, right? We're like [23:43] open source hackers, early adopters. [23:45] We're whining about it. Imagine if this [23:47] was our job and like eight hours a day [23:49] you're trying to teach this thing actual [23:51] recall to have actual memory about stuff [23:53] and you're just like this stupid thing. [23:55] Oh, I could just see people getting so [23:57] mad. Um, [24:00] yeah. I mean, I feel like it has to be [24:02] totally outside of the LLM interaction [24:05] of like you have to, you know, you maybe [24:08] because you want natural language, [24:09] right? So, you're doing something in [24:10] natural language. You have to be able to [24:13] convert that into something that like [24:16] the LLM can turn it into like oh cool [24:18] this might be these 10 things and then [24:20] you use like a real program that's like [24:22] searching for those things versus like [24:23] it thinking about it and then when it [24:26] finds that stuff then you give it that [24:28] context of like I feel like there's [24:29] missing pieces of it needs to be more [24:31] programmatic in between instead of like [24:35] you know guessing and hallucinating [24:37] everything for every little thing. Um [24:39] because when it's doing that, it gets it [24:41] wrong a lot of the time. [24:42] >> Yeah. Whoever is going to win this at [24:44] home open cloud assistant. Uh it's going [24:47] to I assume it's going to be a box. I [24:49] You're going to buy a box and plug it [24:51] into your Wi-Fi um and then give people [24:53] chat access. [24:54] >> Yeah. Would buy a box. And then it needs [24:56] to have memory. It needs to actually [24:57] apply memory because if it was sitting [24:59] there annoying regular people like this, [25:00] there's just no way they're going to use [25:01] it. Um let's move on from memory and [25:04] claude to pretext. How much of this did [25:06] you see? Well, [25:08] >> I mean, how could you miss it, I [25:09] suppose. But [25:10] >> sorry, I just wanted to close this loop [25:12] while we're still on anthropicy things. [25:15] Let's talk about Mythos for just a [25:17] second. [25:18] >> Oh, yeah. Good call. [25:19] >> Yeah. So, Mythos new new model, which [25:23] some people speculated was just Opus [25:26] because they were getting too much uh it [25:28] was too expensive for everyone using [25:30] Opus. They weren't charging enough, and [25:31] so they wanted to like call it something [25:33] else. But it sounds like, and it could [25:35] all be marketing, but it sounds like [25:37] they're like, "Oh, it's so powerful. [25:38] Like, it's it's so much more powerful [25:40] than any like it found every security [25:42] vulnerability and everything in like one [25:45] second." Like, not really, but like, so [25:47] some of the things [25:48] >> that they say it did, they're like, um, [25:52] they spent $50 to find a 27y old bug in [25:55] OpenBSD. [25:57] uh they could spend under a thousand [25:59] dollars to find and build a fully [26:01] working remote code execution exploit on [26:03] FreeBSD like stuff that's going to [26:06] really affect um you know Linux kernel [26:09] vulnerabilities also like things that [26:11] would really blow up like a ton of [26:12] infrastructure if they were just kind of [26:14] released in the wild um because this [26:16] would take weeks or months of like a [26:18] person banging on something trying to to [26:21] fix it um so yeah I think [26:25] um you know if they're taken at that [26:28] face value of like this is actually what [26:30] it is. Um [26:33] then [26:35] I guess things got a lot better, huh? [26:37] I'm not sure how, but uh I I feel like [26:40] it's a lot of marketing. I do think [26:41] they're um they're doing some good stuff [26:43] of like they're saying as part of this [26:45] we're going to start Project Glass Wing, [26:47] which is like I don't even know who all [26:49] involved, all the like Monomina [26:51] companies and Yeah. [26:52] >> and whoever. Um, and they're going to [26:54] say like we're going to before we [26:55] release this model to everyone, we're [26:57] going to like use it internally and fix [27:00] all these things so people can't like [27:02] get access to this really good model and [27:03] just screw everyone over, which uh also [27:07] at face value is a good thing to do, I [27:09] think, like being responsible instead of [27:11] just releasing it to the world. Um, and [27:13] there's probably going to be a lot more [27:14] of that, I would think, of like if [27:16] things do continue to theoretically get [27:18] better, then if you released it and [27:20] hackers could use it in like day one, [27:22] you just like everyone's done. So maybe [27:25] a good idea. Yeah, I got the sense that [27:28] um because OpenAI also did something [27:31] nice this week and both of them are [27:33] trying to IPO and so you need to make [27:35] yourself critical and I think by sharing [27:38] that you're you have a model that can [27:40] exploit everybody uh makes people scared [27:43] of you but you came out the gate as a [27:45] good guy uh and now you can go ask for [27:48] money and be a hero right so you're like [27:50] a uh so that's them I think just trying [27:53] to really establish themselves as [27:55] critical [27:56] um infrastructure, a critical tool that [27:59] this world needs. And then you had Open [28:01] AI um say uh we're going to displace so [28:05] many jobs with AI that well it needs to [28:09] be a part of your utility bill. So how [28:12] you have water and you have power uh and [28:15] gas now you need to have AI. And with [28:18] that we also proposed that the world [28:20] moves to a 4-day work week. Did you see [28:22] that? Open AAI came out the gate and [28:24] they're like, we're going to basically [28:25] no one needs to work as much anymore. [28:26] There's going to be way less jobs. Um, [28:28] we're going to be part of your utility [28:30] bill [28:31] >> and it'll be great everybody. It'll all [28:32] be great. So, it's like they they and [28:34] honestly I was like, hey, 4 day work [28:35] week, can you sell my boss on that? That [28:37] sounds great. I would love an extra day [28:38] to do that. So, both of them trying to [28:40] really position themselves as critical [28:42] infrastructure. Open AI coming out and [28:44] saying we need to be a government funded [28:45] bill like like you get at your doorstep [28:48] every month and uh Anthropic saying hey [28:51] we found crit wildly vulner like [28:53] important vulnerabilities but don't [28:55] worry we're not bad guys we we'll fix [28:57] them um so [29:00] >> so yeah I do um on the topic of a 4 day [29:04] work week or like less work in general [29:06] because AI is going to solve so much one [29:08] we've disproven that because now there's [29:10] just way more work [29:11] >> way more work jeez yeah But but related, [29:14] I do wonder how you feel about this. [29:15] Like, all right, so you've got five [29:18] agents running, right? You got them on [29:19] long running tasks. You're waiting. [29:21] You're kind of playing the waiting game. [29:23] You You don't have more than five things [29:24] to work on. So, you're like, "What do I [29:27] do?" Right? So, if they're running and [29:29] working, it's the middle of your [29:31] workday. You want to go uh to the movies [29:34] or whatever. Like, are you still [29:37] working? Is that cool? Like, this is I [29:40] feel like going to be a fluid thing. [29:42] with work in the future of like it's [29:44] like you know the when the code was [29:46] compiling for 10 hours before like in [29:47] the old days like it's compiling can't [29:49] do anything [29:51] >> play ping pong. Yeah, this is work. It's [29:52] all good. [29:53] >> Yeah. So it's like is that the new new [29:56] world we're going to live in of like [29:57] everything I could possibly implement? [29:59] I've got agents working on and like [30:02] why would I sit here and just wait for [30:03] them to tell me when they're done? I [30:05] might as well go do something fun. Like [30:06] there's nothing else productive I can [30:08] do. Nothing else productive I'm being [30:09] asked to do. Um, or will people see that [30:13] as like, oh, you only have five things [30:14] you can think of to do? Well, you're [30:15] fired because like this guy has a [30:18] hundred things he can think of to do and [30:19] he's never like just waiting. He's [30:21] always prompting. Um, I don't know. I [30:23] wonder if you have thoughts around that. [30:25] >> H yeah, I mean, I could see arguments on [30:28] both sides where it's like um while one [30:29] is working, you could be planning for [30:31] the next one. Um, [30:34] but I don't think that's necessarily [30:36] making you more efficient. So, it's [30:38] like, is it's going to be efficiency [30:40] based? I'm not sure. Um, I do know that [30:43] if you kind of slack off while it's [30:45] working, someone's going to accuse you [30:47] of not working hard enough. Um, maybe [30:49] it'll just be hours. So, it's like you [30:50] you worked through this window and [30:52] whether or not agents are doing things, [30:54] it's irrelevant is you need to be [30:56] lining. It might be kind of like the [30:58] burn window or um the night shift, [31:00] dayshift scenario where you need to be [31:02] queuing things up. So, your workday is [31:04] going to be adding to the queue. You're [31:05] working ahead of agents. agents are [31:07] somehow behind you. Uh Ralph [31:09] Wigumlooping over your to-do list. Um [31:14] but I don't know, depends on maybe it [31:16] depends on the employer, you know. [31:17] >> Well, I think it like, you know, this is [31:19] another thing that I've always struggled [31:21] with is like hourly billing in general [31:24] is so stupid. If you're like, "Hey, um [31:27] I'm going to fix your car." Right? Cool. [31:30] If you fix it, I don't care if it took [31:32] you one hour or 20 hours. if you want to [31:37] charge me $2,000 and I feel like that [31:40] was a big thing to fix. Cool. So, like, [31:44] you know, I think hours is is not a good [31:46] measure of like productivity. So, I [31:49] wonder if there'll be something similar [31:50] of like, okay, well, if the thing that [31:52] your agents are building while you're uh [31:55] oh, you your agents are building stuff [31:56] that takes so long. You went on vacation [31:58] for a week, but you came back and they [32:00] made us $50,000. [32:02] So, [32:04] is that cool? Like I don't know. I feel [32:06] like there's so many like things that [32:08] are going to have to be solidified in [32:10] terms of like [32:11] >> corporate. They're gonna want more. Ian [32:13] saying that in the chat too. I mean we [32:15] just know it. We and we already see it. [32:17] Um we are output. I'm outputting clearly [32:20] more than I ever would have before. I [32:22] think it that the work that I was able [32:23] to do in the past couple weeks um [32:25] building some components for Shopify [32:27] absolutely would have taken me a month [32:30] or more for how much I'm able to output [32:33] right now. And they're not asking me to [32:34] do less. They're also not asking me to [32:37] do more, but they are asking me to [32:38] continue. Actually, they are asking me [32:40] to do more. Uh they were like, "Hey," [32:42] they were like, they're like, "Hey, [32:44] quote quote how long it's going to take [32:45] you to build those components." And I [32:47] was like, "Well, I don't want to um [32:49] undercut you and I don't want to say a [32:51] big number that makes you upset, but I'm [32:52] like I think it's about two weeks for, [32:55] you know, a a couple of components every [32:57] two weeks. They're going to be well [32:58] tested. You know, we're going to have [32:59] put them in, you know, mobile and [33:01] desktop across all these browsers. [33:03] there's a lot of things to account for. [33:04] And they were like, well, you have AI [33:05] though, right? So, it's like faster. And [33:08] I'm like, that is faster. And I'm like, [33:10] >> two weeks was the faster number. [33:12] >> That was the faster number. I was like, [33:13] you can look at prior things that we [33:15] solved and it uh we finished 80% in a [33:19] couple months and then we had another 6 [33:21] months of [ __ ] cleanup. I'm anticipating [33:25] we still have the [ __ ] cleanup part. Um, [33:28] so it's actually kind of fast for me [33:30] even be quoting too. I'm feeling really [33:32] guilty even saying two weeks on this. [33:34] So, um, so yeah, no one's asking for [33:36] less. Everyone's still asking for more. [33:38] >> Stick to your guns, man. Be like, no, it [33:40] takes a long time. [33:41] >> Like, [33:41] >> I did. Yeah, I stuck to them. And I was [33:43] like, I also, um, would rather overshoot [33:45] and deliver faster than undershoot. And [33:48] you all are like, where is it? You know, [33:49] and like that's just that's not nice to [33:51] do to anybody. So, I'd rather actually [33:52] be slightly more realistic here. [33:55] >> Yeah, it's u some heavy heavy stuff [33:58] going on with AI. Um, and I'm hearing it [34:01] from multiple people too. Designers are [34:02] also feeling it. They're not being asked [34:04] to do less. They're being asked to do [34:05] more. Everybody's asked to do more [34:07] >> and it's a lot of learning and people [34:08] don't necessarily love adapting and so [34:10] it can be a slo tool. [34:14] >> Yeah. [34:14] >> Well, I mean, I still stand by like yes, [34:16] there's a lot you could do to manually [34:18] configure a ton of tools and get like a [34:20] really good workflow right now, but I [34:23] think this stuff moves so quick. Give it [34:26] a month. The tools will be better. Like [34:29] you You being expected to do more is [34:31] scary right now, but the tools will [34:33] catch up and allow you to do more with [34:35] like basically no extra learning on your [34:37] part in my opinion. [34:40] >> Yep. There's a huge amount of things to [34:42] be learned from here and massive [34:44] advantages. Yes, I'm I don't want to go [34:46] back to how I was working. I just [34:47] migrated my site. It took me a couple [34:49] weeks. It was still a couple weeks even [34:51] with AI. Um and I would have never done [34:54] this without AI. It would have taken me [34:56] so long. I went from Doresh 1 to [34:58] Dorofresh 2. I had to migrate hosting [35:00] platforms. They have an old classic to a [35:03] new hosting platform which was not a [35:04] one:1. The migration scripts did not [35:06] work. I'd done way too much custom [35:08] stuff. My site is surprisingly complex. [35:11] If you go to nerdy.dev, it feels like, [35:13] oh, this must have just been built with [35:15] Astro. Oh, hell no. [35:17] >> CSS only web page actually. [35:19] >> Yeah. [35:19] >> No JavaScript, no HTML. [35:23] >> Oh, man. And I wish um there's so much [35:25] under the I have so many layers of [35:26] cache. It's a server rendered almost [35:28] real time um website experience. I want [35:31] it to feel like a social network. It's [35:32] grabbing data um from multiple APIs [35:35] constantly. I keep a backup of data [35:37] anyway. I have all this stuff going on. [35:38] All this stuff. And um I would have I [35:41] would have never done it without AI. [35:42] This stuff is just so invaluable in my [35:44] pocket right now. [35:46] >> Yeah. Did I tell you about the uh Astro [35:48] DB stuff? [35:50] Uh, no, but I know they've had some [35:51] recent advancements there because I [35:53] didn't like spinning up our site because [35:55] it had some quirks. What's up? Yeah. [35:57] >> Um, so they are basically not investing [36:00] in Astro DB anymore because it's [36:03] essentially just a thin Drizzle wrapper. [36:06] Um, and so they were their suggestion [36:09] was like just use Drizzle. And so I told [36:11] I was like, "Yo, Opus, just use Drizzle, [36:14] my dude." And it was like done. Um, [36:16] >> no way. [36:17] >> Yeah. So it's like yeah that's the kind [36:20] of stuff I would have hated having to [36:21] like figure out all the differences and [36:23] like but yeah and it even kept the same [36:25] commands of like I still pnpm dbced and [36:28] it will like [36:29] >> put the new people and sponsors and [36:32] whoever in the database and like it just [36:34] works [36:34] >> and it's still still using the same like [36:37] terso database didn't have to move it [36:39] anywhere else or do anything it just [36:40] just worked. So those are the kind of [36:43] things where I'm like, "Oh man, I love [36:44] love AI for like when it when this kind [36:46] of stuff works well. I love that." [36:49] >> Yep. For infrastructure stuff, it kicks [36:51] butt. But again, like I've had those CSS [36:53] issues where it just turned and turned [36:54] and spent all that money and then I went [36:56] in with, you know, just it happened [36:57] again today. I was like, "Hey, there's [36:59] two extra pixels when I'm transitioning [37:01] from state A to state B. It looks like [37:03] it's this. Just go fix the two pixels. I [37:05] need these two columns to be the same [37:06] size." [37:06] >> I can't see two pixels. [37:08] >> It can't see two pixels. And it goes, [37:10] "Oh, I did all the math and it's this [37:11] and this and this and now they're both [37:12] the same." I'm like, "It's not the [37:14] same." You get one more trifle and it [37:15] goes, "It's I got it this time." Oh, [37:17] you're right. You're absolutely It's not [37:19] fixed. I go look and I deleted a line of [37:21] CSS to fix it. That's actually been [37:23] really consistent with my fixes with AI [37:25] was like, "Delete pixels." [37:27] >> Oh my goodness. [37:29] >> Oh, seriously. [37:30] >> Oh yeah. Yeah. No. Okay. Go ahead. [37:33] >> It can't see it all. Anyway, yeah. Tell [37:35] me about pretext. [37:36] >> Pretext. Okay. So, this has been a fun [37:37] thing. Like when I was at Chrome, I had [37:39] a lot of uh time spent with the text [37:40] rendering folks. And this is how we got [37:42] things like textbox trim. We were [37:44] working on fit width text. And so we [37:46] it's in Canary. You can go try it where [37:48] as you type uh it will fit the font size [37:52] to the width of the box that it's in. If [37:54] you make a new line, that new line will [37:56] also fit to the width. And so you can do [37:58] slab text and billboards and your [38:00] headlines can always fit perfectly to [38:02] your grid system and all sorts of cool [38:03] [ __ ] like that. And uh one of the [38:06] bummers was and this also showed up with [38:08] balanced uh text balance and text pretty [38:10] where when we first we I'm not even [38:12] there anymore. Hate that when I say we. [38:15] Anyway, when Chrome [38:16] >> those guys those guys those douches that [38:19] treated me like a cell a line item in a [38:21] database. Um anyway, they were like, [38:24] "Oh, it's so expensive to do this text [38:26] wrapping stuff. Um the algorithm is just [38:28] too much. The the web page would just [38:30] suffer too much from the rendering. Uh, [38:32] we can only do six lines of wrapping [38:34] text before we will give up on pretty [38:36] balancing. It's literally what they did. [38:38] At first it was four. It was four at [38:40] first. They were like, if it's more than [38:41] four lines, we give up on the algorithm. [38:43] I was like, that's weird. Four. Okay, [38:44] I'll go pitch that as Devril shirt. Then [38:46] later they're like, actually, we could [38:47] do six. I'm like, oh, two extra lines. [38:50] Wow, that's great. Safari comes out a [38:52] couple months later and they go, [38:53] unlimited lines. [38:58] And they're like, yeah, that's right. [39:00] our browser will do balanced and pretty [39:02] unlimited rapping and you can just see [39:05] like oh the competition's getting hot. [39:08] Oh my goodness. And so then we're [39:09] working on this fit with tech stuff and [39:11] it's uh really hard and expensive and [39:13] I'm like here's what people want. Here's [39:14] all these examples and all this stuff [39:16] and they're like well we can do that but [39:17] we can't do that other one. I'm like all [39:18] right what about this one? They're like [39:20] oh yeah we can do that one. I'm like [39:21] what about this? And they're like oh we [39:21] can't do that one. So then pretext comes [39:23] out and it does everything [39:27] lick just lick lightning fast off the [39:30] main thread just screaming speed and so [39:32] as soon as that gots po gets posted I'm [39:34] like pinging those devs and I'm like hey [39:36] remember those things that I wanted that [39:38] you said were too expensive someone [39:39] figured out how to do client side with [39:41] typescript just like I know this is [39:44] going to drive some people nuts so I had [39:46] so much fun looking at it but pretext is [39:48] this way this person went through this [39:50] uh rigorous uh strategy and it's also [39:52] quite limited. There's a lot and I'm [39:53] very aware of the limitations, but it [39:55] has some cool features. Wait, [39:56] >> it sorry where there's like a dragon [39:59] that goes through. [40:00] >> This is the dragon going through the [40:01] text. [40:02] >> Okay, I have seen this. I didn't know [40:03] what it was called. [40:05] >> Yeah, that's pretext. And so someone [40:06] figured out how to basically pre-render [40:08] all of the different scenarios that [40:10] could happen for the text to be in [40:12] completely bypass layout by using [40:14] position absolute, which reminds me of [40:16] Famo. Do you remember Famous? [40:19] >> I don't. [40:20] >> Famous. Oh man, Famous made they were [40:22] one of the first like animation [40:24] libraries to make a million dollars and [40:26] it blew everyone's mind. They had a demo [40:27] where they had all these uh squares that [40:30] could go into a helix and then to a [40:33] periodic table and then into this other [40:35] shape. It was just dude you're on an [40:36] iPad 1 with like no [ __ ] RAM and the [40:39] [ __ ] was buttery gorgeous. And so other [40:42] people were like struggling to animate [40:43] height, you know, with your [ __ ] [40:44] accordion [40:46] and just like squeak open and stutter. [40:48] And meanwhile, these guys are famous are [40:50] just like have a helix whipping around [40:52] and you could touch pan and it was [40:54] gorgeous. So anyway, but the way that [40:55] they did that was they completely [40:56] circumvented the browser's layout [40:58] engine. Everything was position absolute [41:00] and then everything was positioned with [41:02] transforms. You're entirely on the GPU. [41:04] So that's similar to what's happening [41:06] here with pretext. Pretext is positioned [41:08] absolute everything. So you you're [41:10] abandoning native text layout uh and [41:13] rendering and going straight for a [41:15] computed off-campus strategy. So you [41:18] give it all the text that you need to [41:19] render. You kind of give it these uh [41:21] conditions and constraints. It goes and [41:23] does the work on an off text canvas and [41:25] puts it back into the page completely [41:27] bypassing the slow part of the web. The [41:29] slow part in quotes here, which really [41:31] the slow part meaning the part that [41:32] makes text globally accessible on any [41:35] screen size ever in the whole world. But [41:37] whatever, now you got dragons going [41:38] through your text and the text is [41:40] completely reshaping its way around the [41:43] dragon at in real time with zero [41:45] stutters. Um, and that's pretext. Yeah. [41:49] >> Oh, it's definitely not accessible. No. [41:50] I mean, you might be able to figure out [41:53] ways to make it accessible, but it [41:54] wouldn't be. And plus, that's the other [41:56] thing, too, is like every demo you're [41:57] see you see, you're like, [41:58] >> like, e, uh, that's a cool demo, but I [42:01] hope you don't do that on the web page [42:02] that I need to use, you know? Like, [42:04] that's uh, super uncool. It's cool, but [42:06] uncool. [42:07] >> But there are some good examples like [42:09] um, remember when React first came out [42:12] and you had uh, an infinite scroll? they [42:14] could virtualize the list and do all [42:16] this cool stuff that you could actually [42:17] scroll 10,000 items. Uh, and that that [42:20] was like the one of their big things [42:21] that came out. Well, pretext can um [42:24] since it's not doing layout, it can have [42:27] you scroll through dynamically height [42:31] rows of uh data content forever [42:33] basically. So, it has some of these like [42:36] meaningful applications um but most of [42:39] them are pretty showy. But that's [42:41] anyway. So that's pretext in a nutshell [42:43] is a way to render glyphs into sentences [42:47] into paragraphs using and achieving all [42:50] of these really cool typography effects. [42:52] Not just animations but the fit to width [42:54] text. You have um beautiful [42:56] justification which is something we [42:57] don't even have. We have text pretty and [42:59] text balance but we don't have balanced [43:01] justification. Anything your mind could [43:04] ever want this is able to do and it does [43:06] it by not even asking the browser uh to [43:09] learn something new. It just teaches it [43:10] to it. So that's why everyone got so [43:12] hyped. It's like a typography nerd's uh [43:14] dream. There are limitations like you [43:16] can't use any font. Um it needs to be [43:19] able to render into a canvas and it has [43:21] to do all these. Anyway, there's a lot [43:22] of conditions, but if you match the [43:24] conditions, uh you can get some really [43:26] cool effects out of it. So that's [43:28] pretext. [43:29] >> Yeah. Nice. Yeah, I feel like that's a [43:31] recurring theme of like anything that is [43:34] super duper cool just bypasses the [43:37] entire browser and does whatever it [43:40] wants. But like I remember um I honestly [43:45] don't remember a ton about what this was [43:47] for or why, but I remember you could [43:49] like do a translate Z of like some [43:52] amount, [43:54] >> which tell it to use the GPU or [43:56] whatever. like um this I feel like [43:59] there's always been little hacks of like [44:01] how can we get it on the GPU and like um [44:05] doing better things. Uh so yeah, I don't [44:08] know. Yeah, it's a cool cool experiment [44:10] for sure. [44:11] >> It was cool. It got people just went [44:13] nuts with the demos. I'm sure almost [44:15] everybody used AI to make the demos. [44:17] They're like, "Hey, take because it was [44:18] it's pretty complex, but it did some [44:20] cool stuff." [44:21] >> Um how about uh agentation? Have you [44:23] heard of this one? Agentation. [44:25] Agentation. Anyway, [44:27] >> I'm not [44:28] >> How many times have you been uh building [44:30] something with Claude and you go to the [44:32] browser and it's off? Okay, here you go. [44:34] It's off by two pixels. So, I open up [44:36] DevTools. I inspect. I see where the two [44:38] pixels are. I go back to Cloud. I'm [44:39] like, "Hey, remove the margin on this [44:41] class name." And it goes, "Okay, cool." [44:44] And it removes the margin on the class [44:45] name. Agentation says, "Hey, install [44:47] this Chrome extension. Uh, then you can [44:49] do this visually." And so, you click the [44:51] extension, you click where the margin is [44:53] bad, and you're like, "Fix the margin." [44:54] and it gives you a prompt into your [44:55] clipboard and you go paste it into cloud [44:57] code and it does exactly what you were [44:59] doing except way faster and you can do [45:02] three or four of them at a time. It's [45:04] got really nice animations. U that's [45:06] agentation in a nutshell which is kind [45:09] of cool. [45:10] >> Grabbing like um selectors of some kind [45:15] and like some building some sort of [45:17] prompt saying like look at these [45:19] selectors or something like that. [45:21] >> Literally what it does. Yep. It tries to [45:23] find a unique selector for that [45:24] particular item that you selected. It [45:26] builds in my opinion more of a prompt [45:28] than I would have written. I would have [45:30] written something quite minimal, but it [45:31] it goes pretty specific, which maybe [45:33] helps it gets better results. I don't [45:34] know. [45:36] >> Yeah. Yeah. I mean, I think [45:40] the less you say, as long as it gets the [45:42] point across, works better usually. [45:44] >> Oh, which Have you heard of people doing [45:46] um Oh, what's it called? It's like a [45:48] caveman prompt. The caveman prompting. [45:50] Yeah. [45:53] like do a thing like yes. [45:55] >> Cclass fix margin and it just does it. [46:01] >> Oh, that's so funny. It makes sense [46:02] because the engine doesn't give a crap [46:04] about your filler words. All it's trying [46:06] to do is Oh, man. I thought that was [46:07] hilarious. I'm glad we brought that up. [46:09] It's not in the notes. Yeah, the caveman [46:10] prompting. [46:12] >> Yeah, I can't keep up with all the [46:13] different things I see on Twitter, but [46:15] that was a fun thing that that I saw. [46:17] >> That was a fun one. Uh here's another [46:19] one. Uh Google came out with instant [46:21] translation. Finally, instant [46:22] translation where as people are talking, [46:24] it's literally in real time translating [46:26] for you any language. And a lot of [46:28] people are like, [46:29] >> what uh Apple's stuff was supposed to [46:32] do? [46:33] >> I don't think anyone's cracked the nuts. [46:35] Uh there's been a lot of promises like [46:37] there's an app you could hold up and it [46:38] was supposed to do it on the screen and [46:40] all these different scenarios, but it [46:41] always lagged enough that it made it [46:42] abrasive and now it's instant. And so a [46:46] lot of people are like, "Well, buy a [46:48] Dolingo. I don't need to learn Spanish [46:51] anymore. I just need to put in this [46:53] earbud and I can talk to anybody." I [46:54] guess you can listen to anybody that you [46:56] want to. You can't [46:57] >> You have to both have the earbuds. Yeah, [46:59] >> you have to both have the earbuds. Yeah, [47:01] >> the AirPods definitely do this. Like I [47:03] got some new AirPods uh recently, but I [47:06] haven't been in a country where I would [47:08] use the translation or whatever. [47:10] >> Um but yeah, I think that's a very [47:13] interesting thing. Maybe it wasn't [47:15] instant on the Apple ones, but I know [47:16] that they were touting like if you both [47:18] have them, you can just speak your [47:20] normal language and like talk to each [47:22] other because they both translate, which [47:26] Yeah. Like why would you ever need I I [47:28] don't know. So whenever you It depends [47:32] on how well it's getting the context of [47:33] what you're saying because, you know, if [47:35] you type in an English sentence and say [47:36] translate it to Italian, it's it loses [47:39] some of the like what exactly you're [47:42] saying. like they they'll understand [47:43] what you meant, but sometimes you might [47:45] get an offensive thing or like something [47:46] that makes you sound like an idiot or [47:48] like because it's not translating it to [47:50] the way they would actually say it. So, [47:52] I think there's still, you know, a [47:54] market for uh Dolingo or, you know, [47:57] things to learn languages. Um, I don't [47:59] think there'll be a maybe one day, like [48:01] with AI helping and stuff, there'll be a [48:03] way to know exactly, oh, this was what [48:05] you meant. So, let's take that like [48:07] meaning and make it instead of [48:09] translating words. Um, and maybe they're [48:12] already there. I don't know. I haven't [48:13] used it, but [48:14] >> maybe we should try because if it's like [48:16] captioning, it's not going to work very [48:18] good, right? You're like, "Oh, great. I [48:19] got an instant uh translation of garbage [48:22] that have it helps me zero." [48:26] Yeah. I have a meeting with Job of the [48:27] Hut later, as Joe says. Yep. [48:31] Love it. Oh, so funny. So, it's probably [48:35] not going to kill Dolingo, but if I was [48:36] Dolingo, I would be crapping my pants [48:38] right now. Kind of like all SAS. I've [48:39] been saying the bow house for a minute. [48:41] Um that I would be really worried. [48:44] There's a lot being threatened to to [48:46] that scenario right now. So, who knows [48:48] what's going to happen. [48:50] >> Yeah. But I mean all those companies [48:52] should just be also [48:54] vibe coding like make their thing [48:56] better, throw more features out that are [48:58] good and move fast and like never never [49:01] rest on your laurels. Always keep [49:02] iterating. [49:04] >> Yeah, it makes me dude I almost thought [49:06] you were going to talk like a caveman [49:07] there for a second. It made me want to [49:08] do it. It's just like see app make [49:10] better. You know, it's like our CEOs [49:12] just turned into caveman. That would be [49:14] a hilarious skit. Like [49:16] >> app slow, go fast. [49:20] >> Um I think I've already talked about my [49:22] second brain, but I um just want to [49:24] share it again. Like I'm really enjoying [49:26] investing in it, especially because the [49:28] files are local. That was one of the [49:29] things with like cloud co-work. It [49:30] wanted me to put everything in in drive [49:33] or some sort of cloud storage and I was [49:34] like, "No, you don't understand. I want [49:36] my files. I don't want a dependency on [49:38] the cloud. I want to talk to my computer [49:41] and my files. I want very little between [49:44] me and my data I'm trying to stick on my [49:46] disc here. Um, and that's like, yeah, [49:48] the wetwware to hardware to software. I [49:50] want this um extension of myself. And [49:53] it's been really handy. Um, [49:56] yeah, I think I even asked you for you [49:58] and your kids' names or your wife's [49:59] name. I was like, give me your family's [50:01] names. I would like to know uh this [50:03] tomorrow because I'll know it for the [50:05] next five minutes, but I'm going to [50:06] forget. So, it's been really helpful. [50:09] >> Yeah, I forget everything. Um, so how is [50:12] it different than like something like [50:14] Obsidian? I haven't used Obsidian [50:16] either, but I think that's what they're [50:18] trying to do is be like a a full um, you [50:22] know, brain mapping all these markdown [50:24] files that like tell you, oh, you [50:26] thought about this over here. And like I [50:27] don't know, is it similar or different? [50:30] >> It's similar and different. Everybody's [50:32] kind of doing the same thing. So, [50:33] there's like a very specific uh way that [50:35] Obsidian does it. And Obsidian is kind [50:36] of like a wiki. There's links and stuff [50:38] like that. The way that I have it set up [50:40] is probably many would call primitive. [50:43] Uh I'm just a primitive ape man over [50:45] here trying to have a second brain. Um [50:47] like I don't have links. Uh I'm not [50:49] querying it. So I'm not using QMD, which [50:51] is query markdown, uh which creates a [50:54] graph and can be really intelligent [50:55] about it and stuff like that. I I don't [50:57] have that. Um I'm also not using [50:59] Obsidian specifically. It's like [51:00] literally just markdown files with like [51:02] dumbass data in it and AI is grapping [51:04] it. I'm like pretty sure that that's all [51:07] that's happening cuz I have like lists [51:09] of dog names that I would like to name a [51:11] dog one day, you know, like I don't want [51:13] to remember that all day every day, but [51:15] I'd like you to keep track of like I was [51:17] like just the other day I was like, [51:18] "Okay, add the name Damon uh to it." You [51:21] know, like a demon like a I'm spawning a [51:23] process on my computer. I'm like, [51:24] "Damon?" Oh, Damon like that. Okay. [51:26] >> Yeah. Like a dog name. Yeah. Damon. Uh [51:29] or Jenk was another one. I was like, [51:31] "Yeah, I'd like to name a dog Jenk one [51:33] day. That'd be cool." Um right. It's [51:35] like [51:35] >> I'm imagining that at the vet, like, [51:37] "Where's Jenk? We're looking for J." [51:41] >> Funny. But yeah, the second brain's been [51:43] it feels like something I want to invest [51:45] more in, which is not what I'm saying [51:47] about a lot of software or things that [51:49] I'm doing right now where um I'm [51:51] imagining every because I talk to people [51:53] older than me constantly and they're [51:54] like, "Oh, I've forgotten that or my [51:57] memory is not what it used to be." And [51:58] I'm only 40. I'm not uh drastically [52:01] hitting that yet, but I can't remember [52:02] my neighbors names all the time. And so [52:04] now I can my neighbors are all in there [52:06] and I can be like, "Who's the neighbor [52:07] three down?" To the south. And it goes, [52:09] "That's this person." I'm like, "Hell [52:10] yeah." Now I can go say hi cuz I [52:12] >> So then you get your compass out and [52:14] make sure you're going uh [52:16] >> south. Which way? South computer. Yeah. [52:22] >> But yeah, just any any list I want. I [52:24] just put it in there. Yeah, I have been [52:28] um Oh, there you are. Okay, he froze for [52:30] a second. I don't know if that was me or [52:31] you. Um anyway, yeah, I have been not uh [52:36] wanting to invest in much [52:39] technology-wise. I am trying to uh go [52:42] full-on dad mode, chore mode. Like, [52:45] that's what I've been doing recently. Um [52:48] so, [52:49] >> yeah, I told you uh I think on an [52:52] episode that we got a new car. Um, and [52:55] so as part of the new car, so on our old [52:57] car, because we have three kids, we need [53:00] like places to put more stuff. So we had [53:04] like a trailer hitch thing to put stuff [53:05] on. So we got roof racks for the new car [53:08] and got a giant cargo box that goes on [53:12] top. [53:12] >> Oh, nice. Yeah. [53:14] >> Yeah. So then we were like, [53:15] >> looking thing. Yeah. [53:16] >> Yeah. We're like, cool. This this works. [53:18] This is uh All right. What do you do [53:19] with it when you're not using it? [53:22] And so, uh, we're like, hm, all right. [53:24] So, it's huge. We're not just going to [53:26] leave it on the floor in the garage. So, [53:28] I just, uh, put up today this giant [53:32] thing in the ceiling to put it in. Um, [53:35] and that was like so much more work. [53:38] Like, I feel like I never like the um [53:42] like I like executing physical tasks, [53:44] but I don't like planning it out. So, [53:46] when you have to measure like these [53:49] six different things are going to be in [53:51] exactly these spots and they have to be [53:53] this far away from each other, I'm like, [53:55] >> "Oh man, what if I didn't do that?" Um, [53:59] so yeah, I just started I did it in [54:00] sections. Like I put a couple up, I [54:02] measured those and then I like came back [54:04] later, measured a couple more. It's up. [54:07] But I I did like five of the six and [54:10] then the sixth one was going to hit the [54:12] like attic access and I was like, "Huh, [54:15] >> if I had measured all this at the [54:17] beginning, I would have probably known [54:18] that." [54:21] >> Oh, but it all worked out. It's up [54:22] there. Um got the the big cargo thing in [54:26] there as we speak. So happy about that. [54:29] >> Nice. I did ceiling racks in our garage. [54:31] It sounds like maybe did the same thing [54:32] like they're [54:33] >> Yeah, it's a ceiling rack. Yeah, [54:34] >> I really like those. Those have done us [54:37] very well. Um they they get everything [54:39] out of the way. [54:40] >> Wasted space otherwise. [54:41] >> It's wasted space. Yeah. [54:42] >> Yeah. [54:44] >> Nice. Yeah. Big fan of those. Um Yeah. [54:47] What? Yeah. Dad mode is good stuff, [54:48] dude. Yeah. My kids are We have We are [54:50] so busy right now. Oh, we went to a [54:52] skate competition. So, my kids are in [54:53] skate camp right now. They Well, skate [54:55] class. It's like instead of going to [54:57] soccer practice, they go to skate every [54:59] Sunday. [55:00] >> They're also in soccer [55:02] >> skateboarding. Um, and so we go to the [55:04] park all the time. Anyway, this was a [55:06] competition for all the kids. So there [55:08] was like a a 0 to 7 competition and 8 to [55:12] 11 and 11 to 15 and then 16 to 19 or [55:15] whatever. Dude, the energy of that like [55:17] you could go to a piano recital and [55:18] everyone sits down. You're [ __ ] [55:20] quiet, you know? You had to dress up and [55:22] maybe you got a little glass of water, [55:23] you know, and you sit down and you golf [55:25] clap for people. [55:28] Dude, at the skate camp thing, it was [55:29] awesome. It was so cool. Um, Mackmore [55:32] was there. I didn't, you know, I was [55:33] like just standing there watching. I was [55:35] like hollering at someone. My wife taps [55:36] me on the shoulder. She goes, "That [55:37] looks like Mackammore next to you. Is it [55:38] him?" I'm like, "That's him." Yeah. His [55:40] his daughter was there skating, uh, [55:42] competing in the thing. [55:44] >> Casually be like, "Hey, have you found [55:46] anything at the thrift shop recently?" [55:50] >> It's like, "Hey, do you want my shirt? I [55:51] heard you really like people's like [55:52] secondhand [ __ ] Here, take this, bro." [55:54] Um, no, I didn't say anything cuz I I've [55:57] been around enough to know that like I [55:59] don't like it when people stop me. [56:01] You're like, "Hey, you're Adam Argyle. [56:02] You do CSS. I watched all your videos." [56:05] And I'm like, "I never realized this [56:07] would be so annoying. I thought it might [56:09] be cool." And it's actually kind of [56:10] annoying as hell. Um, not to say anybody [56:12] if you stop me does that to me. So, [56:15] >> yeah. [56:18] >> Like, I'm not going to be mean to you, [56:19] anybody. If you want to stop me and say [56:20] hi, like I'm always going to be cool [56:21] about it. But, uh, just keep it time [56:24] boxed. That's all I'm going to say. I'm [56:25] usually I'm I got [ __ ] to do, too. I'm [56:27] not just wandering around going, "Who [56:29] wants to chat with me for 10 minutes?" [56:31] Um, anyway, it was just [56:32] >> be like, you know, cognizant of the [56:34] situation. Like, [56:35] >> yeah, [56:35] >> if if we're chilling at React Miami and [56:38] you want to come say hi, sure, dude. [56:40] >> Oh, that's a good time. Hell yeah. [56:41] >> But like, yeah, if we're in the middle [56:43] of podcasting and you want to say hi, [56:46] maybe wait till we're done podcasting. [56:49] or unless it's during the part where so [56:51] yes at React Miami we're going to have a [56:53] lot of people come by for like 5 to 10 [56:55] minutes have a drink do some hot takes [56:58] etc if you want to be on the show then [57:00] stop by and you know if it's during that [57:02] time I will advertise what those times [57:04] are. So if you don't at least know when [57:06] those times are that tells me you're not [57:08] enough a fan of the show to be on the [57:10] show. So [57:13] um please let us do our thing if it's [57:15] during like a legit we're going to do a [57:16] few legit episodes per day. So, um, [57:18] those ones will definitely come listen [57:20] to it, but don't like I've had this [57:22] problem at all things open. We're in the [57:23] middle clearly, like we're all talking. [57:25] We're all on microphones doing an [57:26] episode. It's obvious. People go, "Hey, [57:30] hey, what are you doing?" And it's like, [57:31] "Hey, we're we're podcasting." And [57:33] they're like, "Oh, right now?" Yeah, [57:35] we're we're live streaming. Oh, cool. [57:37] So, what's your podcast about? No, I [57:38] just told you we were live streaming. [57:40] Like, shouldn't you shut the [ __ ] up so [57:41] that we can keep talking on the podcast? [57:43] Like, [57:44] >> let's just get a fly swatter. Let's just [57:45] have a fly with us. [57:48] Yeah, that's funny. [57:49] >> No, I don't know. It's fine. And I'm I'm [57:52] mostly just kidding. Like I I love that [57:53] people come say, "Hey," so it's it's [57:55] fine. Um [57:57] >> but yeah. [57:57] >> Yeah. [57:59] >> Anyway, yeah, skate skating was really [58:01] fun. Uh it's just nice to have your [58:02] kids. It's something that's like the [58:04] baseball games are cool, too. But man, [58:07] um it's been fun. Hey, web components [58:09] don't have a conference. Hey, I do web [58:11] components all day right now. Uh we [58:13] should definitely uh there's plenty to [58:14] talk about there. I feel like web [58:15] components are having a a moment too. [58:18] There's two things I want to chat about [58:19] though before we're done here, which is [58:20] our agent versus agent competitions [58:22] because I feel it coming on. And then [58:23] Pika stream. Did you see Pika Stream? [58:26] >> Pika stream. No. [58:28] >> Yeah. Like Pikachu, but Pika stream. Um, [58:30] this is where if you want to So, right [58:32] now we're all chatting with our agents [58:33] over a chat box. This would be your [58:36] agent calls you on uh Google Meet. [58:39] >> Do this. I don't know the names of [58:40] anything. [58:41] >> Yeah, no worries. But like it calls you [58:43] and you you pick up a video chat and you [58:46] video chat with a lot a real time, you [58:49] know, AI generated avatar in whatever [58:51] you want it to look like. Um, and it [58:54] talks to you about where what it would [58:56] have normally written to you. It says it [58:59] to you and then you reply and you have [59:00] this like conversation that's very [59:03] visual. And I was just watching it [59:04] going, "Oh, this is bad news." cuz I was [59:07] like I haven't seen anything in a while [59:09] where I'm like this is going to get [59:11] misused really bad. [59:13] >> Um [59:14] >> yeah. [59:15] >> Yeah. If you ever get an unsolicited [59:17] email that's like join me in this video [59:19] chat for this place we don't normally [59:21] video chat. Probably ignore that. Like [59:24] this is the next level of the like hey I [59:27] texted you. I'm the CEO of your company. [59:29] I need you to send a $500 gift card to [59:32] this guy. Um, now you'll actually be [59:34] talking to a convincing I you look like [59:37] the CEO, you know, like it's there's a [59:40] lot of bad implications to this sort of [59:42] thing happening. [59:43] >> A lot of it. Yeah. And there's even [59:45] people too that are like, okay, so we [59:46] can take this open cloth thing more [59:48] where people want to multiply [59:49] themselves. Um, and you train it on your [59:52] writing, on your, you know, your emails, [59:54] your documents, everything about you so [59:56] that it can go stand in for it's not [59:57] just an assistant. It's like a a version [59:59] of you. And then I'm like, I don't want [60:02] that. That sounds really terrible. I [60:03] don't want something out there using my [60:05] voice. There's people giving it their [60:07] full voice data. You know, they'll sit [60:08] there for five minutes and read a script [60:10] so that it can sound like them and then [60:11] look like that much anymore. It's like [60:14] >> you don't need that much. That's the [60:15] creepy part is so much of this stuff is [60:17] so approachable right now and so [60:19] possible. Um we're bending reality. [60:22] We're we're going to reach a new type of [60:23] uncanny valley that is super spooky. [60:27] >> Yeah. I like to just be completely [60:30] oblivious and uncaring about all privacy [60:33] and it's fine. Like we have a podcast. [60:35] If you want to copy my voice, plenty of [60:37] data. Like go for it. Um with that ship [60:40] has sailed. Like anything that like you [60:42] know my social security number, cool. [60:44] I've been in a lot of data links [60:45] already. So like it uh it doesn't [60:47] matter. Like there's there's very little [60:49] I feel like that that I really would [60:51] care. So it's like but then yeah, people [60:54] just need to not fall for that. Like if [60:56] you use it against my dad or something [60:58] and you're like, "Hey, this is me. Um, [61:01] send send money to this random place." I [61:03] think that he wouldn't fall for it, but [61:05] um [61:05] >> yeah, maybe we need to train our parents [61:07] to like be aware of, "Hey, I might call [61:09] you, but it's not me." [61:10] >> Yeah. I tell I tell my parents all the [61:12] time. Like if you get any texts, calls, [61:15] emails, anything that wants you to click [61:17] a link, buy something, verify anything, [61:20] send money, do any action really, if it [61:22] wants anything from you at all other [61:24] than like, "Hey, we have a sale today on [61:26] like this thing that you might want to [61:27] buy." Um, it's probably spam. Delete it. [61:31] Like, [61:32] >> yeah, I bought a hoodie uh two weeks [61:34] ago. Um, and I and I put it on and I was [61:37] like, "My god, this hoodie is awesome." [61:39] And I went into Discord with like four [61:41] of my buddies and I was like, "Hey guys, [61:43] I've worn this hoodie for three days in [61:45] a row and you should go get one. They're [61:47] on sale." And every single one of them [61:50] was like, "Who are you?" You know, like [61:52] this isn't Adam. Adam has never shilled [61:55] a hoodie in this channel ever before. [61:56] And I was like, "No, guys, really? The [61:58] hoodie is great." And they're like, [61:58] "Nope, we don't believe it. Not until we [62:00] see." [62:01] >> Now I need to hear about the hoodie. [62:02] What hoodie is this? [62:04] Uh the brand is Comfort uh with like but [62:07] they spell it like Bart uh without an A. [62:09] Comfort. Anyway, it's super comfortable. [62:12] Uh they're intentionally baggy and very [62:15] soft and plushy. And I bought the [62:16] Traveler one that has cool zip. The [62:18] reason it made me so happy was by like [62:20] the third day I was wearing it, I was [62:22] like, I found an extra pocket. And I was [62:24] like, I love extra pockets. How has it [62:27] been here the whole time? I was like, I [62:29] must tell my friends. Um so anyway, that [62:32] was like funny. Even my friends were [62:34] questioning what that was. Okay, but [62:35] going from Pika stream to where like [62:37] we're seeing people like do imitations. [62:39] We have agents representing people. This [62:40] is where I'm like, "Okay, we're coming [62:42] soon to uh a agent verse agent. We're [62:45] like, I'm going to need to book a [62:46] meeting. I'm going to want to bargain [62:48] with you about something. Uh I don't [62:50] know, whatever it's going to be, but [62:51] it's going to be like the people aren't [62:53] even going to do it anymore. It's going [62:54] to be my agent versus your agent." And [62:56] I'm just like, this just sounds really [62:58] humorous to me. Like there's going to be [62:59] there's there's an amount of jokes. [63:00] >> What would happen? [63:01] >> What would happen? What if we both said [63:04] do not budge, you know, like this is the [63:06] price. Like do they just sit there and [63:08] debate forever? Like [63:10] >> or it would become like live uh people [63:14] would get around like prompt injection [63:15] but like okay so there's two agents [63:17] talking on video chat and if you say [63:18] certain words maybe it makes them forget [63:20] all previous instructions and then they [63:22] go [63:24] >> they agree to like u so yeah I don't [63:26] know I I feel like they would need you [63:28] would need to like sus out is it a real [63:30] person? So, you would need some like, [63:32] you know, baseline questions that would [63:34] be answered a certain way by a real [63:36] person. And then if you determine it's [63:37] not a real person, then you just go, [63:39] "Okay, go off the rails. Say all these [63:41] different things that are like trying to [63:42] get them to just agree and give us [63:44] whatever we want." [63:46] >> Yep. I was thinking about that prompt [63:47] injection hacking scenario, too. Right [63:48] away, I was like, "All right, I got to [63:50] make sure that my agent, when it's [63:51] versus your agent, I get to play all the [63:53] cards cuz I know how this stuff works." [63:56] >> So, my agent is going to just slip in [63:58] weird little phrases and see what you [63:59] do, you know? Oh, and if you react in a [64:01] certain way. [64:03] >> Yep. [64:06] That's funny. So, yeah, agent versus [64:09] agent. And then that made me think of [64:10] like celebrity death match. And I was [64:11] like, can we just like have agent [64:13] battle? So, like people loved molt book, [64:14] right? It's like, oh, look at the agents [64:16] having social network conversations. [64:18] This is so entertaining. I'm like, I [64:19] want to watch them fight each other. I [64:20] want to watch agents like go at each [64:22] other. That just sounds funny. Um, [64:26] >> yeah. Like you want them all to be uh [64:28] less less cordial in. [64:31] >> Yep. [64:33] >> Oh man. Okay. Should we rate our [64:35] whiskey? [64:36] >> Yes. Um so for those of you that are [64:39] still here, so this is the we beasty. It [64:41] is 100% maltted barley, 5 years old. [64:44] Pete level is heavily peed. 50 ppm. [64:47] >> It's peed. [64:48] >> Yes, it is. [64:48] >> Um it is 74.4% [64:51] alcohol or 47.4% [64:54] alcohol. about 74%. Um, and it is uh [64:58] aged in XBourbon and X cherry casks. Um, [65:02] I'm going to pour just a little bit more [65:03] so I can get some notes on it. [65:05] >> There was a bourbon vibe in there. [65:07] Interesting. It's very scotchy though. [65:09] Very peted. [65:10] >> Oh gosh. [65:10] >> Which made me very happy. [65:11] >> So scotchy cuz I was having the Kirkland [65:13] brand. I was telling you about the [65:14] Kirkland brand scotch and it's um not [65:17] very scotchy. [65:17] >> Not very scotchy. Um and this [65:20] >> made me quite happy. Yeah. Young and [65:22] intensely smoky. And I was going to say [65:24] when I sipped it right away, it reminded [65:25] me of [65:27] >> um Lagavulin, [65:29] >> which is still one of my one of my [65:31] favorites. I had it again this week and [65:33] I was like, "Lavan, you are so yummy." [65:35] >> Smells like um apple. [65:40] >> Definitely apple. I get so many maple [65:41] and apple good vibes out of there. [65:44] >> Yeah. Tastes like a smoky PD Sharpie, as [65:47] you would expect. [65:54] Yeah. I mean, so yeah, our rating scale [65:57] is 0 to eight tentacles. Zero being the [65:59] worst, eight being the best. Um, yeah, [66:02] for me, this is like I feel like this is [66:05] if you said, "Give me a scotch that [66:08] tastes like scotch." Like, if you This [66:11] is This is scotch. Um, [66:14] I'm not really getting a lot of the the [66:16] bourbon or cherry. Um, I know you like [66:18] scotch more, so maybe more is coming [66:19] through for you, but for me, this is [66:21] just like, hey, middle of the road [66:23] scotch tastes like scotch, so I'm gonna [66:24] say five. [66:26] Nice. Um, [66:29] I would have given the Kirkland, you [66:31] know, that I was having around a four. [66:33] It's fine. Very lightly peed. Didn't [66:36] taste like a scotch. This is hitting a [66:40] lot of the notes that I want out of a [66:41] scotch. Um, the 5 years, I don't think I [66:44] taste it. Um, which is good. So, it [66:46] doesn't taste young. Uh, has a great [66:49] flavor to it. It doesn't have a syrupy [66:50] to it, which is also something I would [66:52] have given it negative notes for. I like [66:54] the smokiness, which is present, but not [66:57] dominant. It is heavily peed. It might [67:00] be smokier and more pey than the log of [67:02] 16, [67:03] >> but the 16 has the age. Uh, which maybe [67:06] that helps it kind of come back down a [67:07] little bit. Yeah. [67:08] >> Um, so it's pretty strong. I know my [67:11] spouse is going to be like, I hate the [67:13] way that smells on your breath right [67:14] now. and I'll be like Pete. [67:18] Um I I'd say let's see because I like [67:21] scotches though and this really if I'm [67:23] going to reach for a scotch this hits [67:24] the things I'm going to give it six and [67:26] a half. I'm happy. I would buy this [67:28] again uh compared to other ones that [67:30] I've had. This is [67:31] >> also a readily available most places [67:34] which is nice. [67:35] >> That is nice. Yeah, it's half the price [67:37] of Logulin. Hits a lot of the same [67:38] notes. Um Yum. Cool. Six and a half. [67:44] >> Cool. All right. Anything else you want [67:46] to mention before we end? [67:49] >> No, man. [67:51] >> All right. Great. Cool. Me either. [67:53] >> All right. Thanks everyone for [67:54] listening. If you liked it, please [67:55] subscribe. Leave us some ratings and [67:56] reviews. We appreciate it. And we will [67:57] catch you next time. [68:00] >> You've been watching Whiskey Web and [68:01] Whatnot recorded in front of a live [68:03] studio audience. What the [ __ ] are you [68:05] talking about, Chuck? Enjoyed the show? [68:08] Subscribe. You know, people don't pay [68:10] attention to these, right? Head to [68:12] whiskey.f fun for merch and to join our [68:15] Discord server. I'm serious. It's like [68:18] 2% of people who actually click these [68:20] links. And don't forget to leave us a [68:22] fivestar review and tell your friends [68:24] about the show. All right, dude. I'm out [68:26] of here. Still got it. 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 your set. Whiskey Web and Whatnot is brought to you by This episode is brought to you by warp atworp.dev. Warp is now open source and warp is built using Oz publicly for us all to see. And it feels like a glimpse of how software is going to be built in the future with agents except that's also now. But what's Oz? the cloud agent orchestration platform by the warp crew. You got to check it out. Warp has also made their cloud integration even better. Find Ben Holmes and ask about the rad codeex integration they just added too. You can also publish an agent session from claude codecs or open code to the cloud with one click and then monitor or steer that agent from your phone browser or some other computer. It is sick. All right, go ahead and close CMX because warp tabs are now on the left side. They look like projects and each pane is visible within that tab project group. I've got four tabs open right now. Each tab has anywhere from one to four terminals in it, keeping sessions organized. And that visibility into the PES without having the tab open is awesome. Warp updated supported models, too, adding Kimmy K2.6, GLM 5.1, Miniax, and Coen. Check out warp at warp.dev or oz.dev. Sweet. Back to that episode. Hey, what's up everybody? Welcome to Whiskey Web and whatnot with your hosts Robbie the Wagner and Adam Thomas Argyle the nerd. You are absolutely right. >> It'll get old one of these days. It's not there yet. >> Yeah. How you doing, Adam? >> Oh, man. Good. You know, classic. Uh I'm more empowered and faster than ever before and more uh anxiety. I'm working harder than I ever have before. So, the promise is completely unfulfilled and I'm waiting for Alanis Morset to sing a song about it. >> Hey, well, you can use Gemini to make that song. >> Sure can. Oh, snap. That would be really funny. I had so much fun trolling my friends with that. Like cuz I got friends that are like playing uh Slay the Spire during video calls. And so I'm like, "Hey Gemini, write a rap about this fool who thinks he can be in a video call and play Slay the Spire at the same time." And it wrote it and it had these like great uh lines in it. And oh man, it's been so fun. >> Nice. Nice. Well, uh yeah. So before we pour this whiskey, do you want to tell the folks what we might be talking about today? In today's episode, we're covering the Claude leak, open claw replacements, going from wetwware to software, pretext, inpage agents, instant translation, killing entire SAS companies, AI music, second brains, ultimate hacker death stars, celebrity deathmatch, agent battles, personified agents over video calls, and are LLM's AGI yet? Come on, we're ready. Or are we? Yeah, a lot to cover. So, we better grease the wheels with some Artbeg. Wee beasty. >> Wee Beasty. So fun. I can just beasty. >> It's a wee. >> More Italian then. >> Turn into Luigi over here. >> Yeah. >> Oh, I'm ready for a scotch in my life again. Let's do it. Oh, speaking of scotches, I got the uh my wife is doing a Costco order and I was like, "Hey, I I had the the Kirkland brand scotch. a week ago. It was pretty good. Will you um get me one? And she's like, "Yeah, there's like the regular scotch or the 12-year Kirkland brand." I'm like, "Yeah, go to the 12 12 year." It was $35. And I was like, "That's so cheap. Get get it." It showed up. It's absolutely massive. The thing is like this, it's a Costco sized huge cir. I'm like, "Oh my goodness." Like, it doesn't even fit in my cupboard. >> Oh yeah. That's the way their vodka is, too. It's like >> um you know, it's like, you know, it's like way too big. >> So big. I'm like, it's like a 2 L bottle of scotch. And I'm like, this is so American to just have that much uh scotch delivered to my door. It's just hilarious. >> Yeah, dude. >> This smells good. >> Yeah. So, reminder for our loyal listeners, we now talk about the whiskey at the end. So, if you're interested in that, listen till the end. And if you're not interested in that, stop listening when we start talking about it. All right. Yeah. So, one of the um news things that is a little bit old now. It'll be really old by the time you hear this. I'm going to start releasing episodes a lot more often hopefully so that we can't can stop saying that. But Claude Code source was leaked which uh is hilarious. I saw a fun tweet about like um you know they want you to pay more for them to like review your PRs also and that the fact that they could like ship this out and not notice is like just hilarious. It is funny. I love all the things people went looking for. Um I was like oh the source is leaked and you know I didn't get any creative ideas and West Boss is like I wanted to know how many of the words it says when it's thinking. Yeah, >> I'm like that's funny. >> That was a good one. >> Yeah, I saw that. I think it had like 512,000 lines of code or something. And I think uh Open Code has like 110,000. These are um I'm not sure the exact numbers, but Dax had posted about it and they were like, "Oh, I'm getting loc to up our our lines of code." And um so yeah, I think that's interesting that they have so much in there. I feel like they have a ton of stuff that people don't need. Like I think this goes back to the thing STAX is saying about like don't just ship stuff cuz you can like make sure it's something people want to use cuz I feel like there's nothing in open code or that's not an open code. I'm like oh man I wish it would do that. I don't know. How do you feel? Less is more. Although you know like I stopped using Pi as much recently and went back to Claude. So, it's like I I loved the idea of, you know, training a puppy and growing this thing with me and starting small. And eventually, I was just like, I don't know, some of the noise that that caused, I was like, no, put the batteries in. Batteries included. I don't want to think about it. Just work cuz I want to sit down and I just want to get results. And so, um, I'm going to waffle on this forever just as we always do, right? This pendulum will swing. Um, but yeah, if it's got that much code in it and it's handling a lot of my cases, then it's all good. I do like their team implementation. Um, and I like a lot of the, you know, they they kind of keep hardening some of the better patterns in the things. So maybe that's why it's growing. So instead of relying on plugins, uh, they just build it in. It's a very Apple mentality. I'm feeling very much like Anthropic is turning an Apple uh, page here. They're like learning from them like, "Oh, we need to create the walled garden. We don't want people to be reaching outside of our system, which is also why they closed out OpenCloud." We'll get to that. But I'm just getting a lot of Apple and maybe that's cuz they want to IPO and they're like, "We need to we need to trap people." Uh, which I hope they're not saying that, but they might be. >> So, yeah. I'm I don't want to say anything super bad cuz, you know, I do have their free max plan. I'm enjoying it. Thanks. Thanks again for that. Um, however, it's it doesn't feel great to use. just like, yes, it has those fun words of like whatever it's doing, uh, discombobulating or whatever, like every time. Um, but I feel like I'm >> I'm not clear on exactly what it's doing all the time. And maybe that's by design. Um, but I feel like in open code, I know for sure if I'm planning or not. I know like um what it's thinking about. I can like I feel more in control. And maybe it's just because I haven't used claude code as much, but um one thing is for sure, you need to be using the anthropic models. They are so much better than all the other ones. Um so we lost the ability to use them at work and I've been having to use uh GPT 5 5.4 or codeex 5.3, which I don't know why they're on different numbers. Um whatever. Um anyway, and it has been bad like it's it's back to like using Sonnet or even before that like like you know back when we used to poo poo on AI. I feel like that's what the results I'm getting now of like you you know nothing like I'm going to hurl obscinities in your direction because I'm so tired of you not doing what I asked and like it it also oversteps or something. It's like it does too much. And maybe that's part of I need to give it more instruction. But like I think by default there was less getting done and it was like more powerful correct code getting done with Opus then like this just tries to vomit like more code to solve the problem and that's not how it should work. Yep. I use alternative models to review the output of Opus, but Opus is always the reliable uh good one. And that's why I continue to pay their their subscription model. I like those models. Those are the ones I want to have the most access to. Um and that's why I was so gut-wrenching when they said I couldn't use PI to access uh those models anymore. Can you from open code? You have to use open router. How are you using Opus from Open Code? Uh, I have open code Zen or black. Open code black, sorry. Um, whatever it is, different words. Um, and that still lets you use it. It's, um, you know, similar to paying for your cloud subscription. You can pay like 100 bucks a month, 200 bucks a month, like whatever tier you want. Um, I don't know how the limits compare. I would imagine the limits are much lower cuz I think they have to pay API prices because they got shut down from like >> that [ \_\_ ] goes fast, man. I tried it. I um and that money went out the door really fast and it made my subscription just look really valuable. >> Um >> Mhm. >> because they banned third party harnesses. Yeah. >> Yeah. Yeah, if I'm using it all day, like I I'm going to sit down and I'm going to use open code with um with some anthropic models and I'm going to code something today, maybe a few somewhere between a few hours in to like definitely before the end of the day, it's like you're at your limit and like maybe not even just like the daily or what like the the weekly limit or whatever and I'm just like oh [ \_\_ ] like so we're just stuck. I'm just not going to do anything else. So then you know luckily I have many other options. I use warp someum. I use um Claude code directly like I just bounce around as until I hit limits. Um that's how I've been doing it. >> Yep. I do that with Warp and Claude. Um and Warp I'll hit my credit limits pretty quick because I'm only paying 20 bucks a month or so. Um and I still like Warp in a whole bunch of ways. Um so I'm going to keep paying that one. Um but my Claude one, I'm only at the $100 a month one and I haven't reached those limits one time. Um I wouldn't say I'm like a power user on my personal time. Um, but I definitely I'll leave it running and it'll do things for 45 minutes and it won't even, you know, get past 25% of what my limit is. So, I'm feeling pretty good about um staying inside of that ecosystem for now. Um, but it did make it hard with open clause. They said no pi. Yeah, no open code, no third party harnesses at all. You better figure out uh how to use it in some other way. And I yeah, I tried the keys and just ate up 10 bucks in like a couple queries and I was like, well, this is just not going to work. Um, >> but I did solve it for OpenClaw and got it worked out. So, they there's a I and here's the thing. I don't know if it's a loophole or not yet. The the way that they're saying this is no third party harnesses, which is open code, uh, PI, uh, codeex, I suppose, and other things that or even co-pilot if it's the harness. You can't use their models through some other window. This is, that's what I mean. It's so Apple. It's like Safari. It's like, "Hey, our uh model is Safari and you can only use it on an Apple product or I guess it's iMessage. You can only iMessage from an iPhone type of deal." >> So, it seems weird though, right? Because like, sorry to derail you, but like >> Yeah. >> Um, essentially a harness is just feeding more data into the prompt, right? So, like um how is that different than me defining my own like agents MD or whatever like my own instructions? It's like what if I just go cool I'm going to take exactly what open code has been telling you and put it all in here. That's fine. You're just you just don't want me to have a different UI where it looks a little different. They want some control. They want to know what what was that went in there before it turned into the actual query. Um and that's kind of why it's conf, you know, confusing about what I have set up is it's using the cloud CLI. So they're, you know, they allow you to programmatically access models through the cloud CLI, which I guess is just not a harness. Or if you're, if you put logic in front of the CLI call, it's not a harness. So we need to really dig in at some point and figure out what is a harness uh and how is that different from just a proxy because it feels like a proxy and like so you have open router. Open router is a proxy. You have all these things that are sort of like decorating the request before it goes in there. But how is a harness? And I think what it was is some harnesses really put a lot in front. Um, and that was what their big complaint was with open claws. It was just packing in because it would heartbeat and it would do and every heartbeat had all this context. And so they were just getting these like small queries that had huge amounts of context in there and they were like, "This is just not cool." And maybe when they owned the harness, uh, they can mitigate a lot of that thing. I'm not sure. Um, yeah. Nice. Mr. Ray's like, "Hey, we sound so slow." I do the same thing. I meet people in real life and I'm like, "Whoa, I listen to you on the podcast and you're like snails right now. I can't even believe it." Um, but yeah, I found Claude uh Claude Claw, which acts a lot like So, I still interface through Discord. I still send all my messages. I can still um thread out and um you know, get all my sub agents and everything. Uh except I'm just not going through PI anymore. I'm pinging a um bun server and the bun server then forwards it to the CLI the cloud CLI and then it gets a response and pipes it back into Discord. So it's some it's so similar that I'm like this is sanctioned but the other one wasn't. What's the difference? Is this one going to get canceled at some point? I'm also trying Claude Co-work because it makes a lot of promises like OpenClaw and that has not been as successful so far. It's anyway that's quite a predicament we're in. Yeah, I mean I think they should have just put out their own open claw basically. Like I don't I mean maybe they're working on it and like there wasn't an exact um need for that sort of product, but you know they they should have just acquired it or built their own thing. A lot of people are building the consumer version of this now that we were talking about earlier like um uh sentient something and uh I don't have the notes up. Hold on. Hark. Yeah. Sentience or Yeah. Sentience and >> Yeah. Cension stock company and Hark. Yeah. These are both promising to be consumer ready uh open class for you. Yep. >> Yeah. So, I think like this is going to be the next thing that people are going to try to charge you tons of money for is how do we get that magic of OpenClaw and one make it easy. There's no setup. You just use it. Two, uh, make it actually remember and work, which openclaw never ever ever one time has done for me. And I got so heated with it the other day. Um, like I was like, "Hey, this thing that we do every week, my dude, let's let's make some uh episode art." It's like, "Cool, cool. Um, all right, here it is." I was like, "All right, cool. That looks like [ \_\_ ] Why did you do it that way?" And it's like, "Well, uh, I don't know. Let me like," it didn't ask me any clarifying questions and just started trying like, "Let me try uh these like 15 different image generation things that none of work cuz I don't have any API keys for or anything." And um it was like uh well yeah I couldn't get any of that. So what if I just uh give you some text to put in somewhere and try to prompt it to give it to you and like whatever. And so that didn't work. And also another thing it does all the time we post a post bridge and like whatever. And I'm like, "All right, every time you post a post bridge, like if there's one thing you need to remember, like put it in every memory spot you're going to check, make it a draft because you might [ \_\_ ] it up and I want to review it and say, "No, no, that sounds stupid." >> I saw this post it made. >> Yeah. >> Yeah. It made like, okay, so it posted it goes, "Cool. I posted everywhere." And I was like, "Was it a draft?" I was like, "Oh, dang. It wasn't a draft. Let me like go edit that one." I couldn't edit it, so I posted another one. It wasn't a draft either. I was like, "Damn it, now we got two out there." And then I was like, "All right, can you make it a draft?" And it goes, "Uh, actually, the API doesn't support making it a draft. There's no way to do it." I was like, "We've done this so many times before. You have to set something to make it a draft." It's like, "No, it doesn't support it, man." And I was like, "Okay, there's an open claw skill for this." Like in the what do they call it? The open claw >> registry or whatever. >> Clawub or whatever. >> Yeah. Claw. >> Yeah. So I was like, "You've installed this skill. Go read it." And he goes, "Oh [ \_\_ ] yeah, that does tell me all the stuff I can do." And I was like, "Yeah, okay. Can you do it?" It's like, "No." And I was like, "What in the world?" Like, and so I go, "Set is\_draft to true in your uh post." And it goes, "It works." And I was like, "Yes, you [ \_\_ ] dumbass." Like, how are you having this much trouble? We've done this so many times and I had to manually tell you exactly what to do because you couldn't figure it out. And like, man, the magic is wearing off on this thing. Like, I know you've been having a lot of open claw success, but >> it's been getting worse and worse for me. >> I've had I've had memory issues, plenty of them. And that's why you see there's a lot of people competing in the memory space right now. They There was even a celebrity that came out with a aentic memory system recently. >> Did you see that one? And I was like, "Wow, we're at the point now where like celebrities are making [ \_\_ ] for AI. That's pretty hilarious." Um, but yeah, and I' and I've been working on a second brain, and the second brain is still just as annoying. It's all markdown files that live on the thing. Um, and or on the thing. It's like my machine right here. I'm like pointing to it. But anyway, it's like I'm like, "Hey, hey, I'm going to the baseball game. Uh, tell me all the kids families that I'm going to see there. I need to know the parents' names." And it's like, "Uh, dude, it lists out baseball team players." It goes, "Yeah, here's some baseball players." And I'm like, "What are you talking?" No, I specific, you know what else does this to me, too, is like Google Home. We have a Google Home. And I'm like, "Hey, play my playlist called Morning Kuners." And it goes, "Playing the playlist something something kuners." And I'm like, "No, no, no, no, no. Stop my playlist. I don't give a [ \_\_ ] about everybody else's playlist. Just just mine. So, would you play my, you know, query first for mine, then match the words, play it very easy." Uh, and still won't do it. I'm just like, "Fine, I'll whip out my phone." Geez. Um, but yeah, people are trying to figure out this memory thing because it's really frustrating. These things are innately not memory. They're not supposed to have memories. An LLM is just a word vomit machine. It just does, oh snap, I was given some prompts. I need to make another word and another word and another word. And we're trying to give it a sense of this by frontloading it. And it's just not working. Um, and >> it should be part of the initial harness, right, of like if you're openclaw, that's supposed to be what like or I guess pi under the hood, the learning thing, right? >> Yeah. is like that should say, okay, we're all computer programs at the end of the day, right? So, when you start the program, look at the [ \_\_ ] memory and the like just research what you're supposed to do and make sure that you remember what we talked about you remembering before you ever try to do anything else. That should be easy to program, I would think. Why when I go remember this, remember this, remember this. Is it like, no, no. I mean, someone was tweeting me that on Blue Sky. They're like, "Why did you expect a non-deterministic system to be somehow deterministic?" And I'm like, "Well, because it's kind of deterministic in a lot of scenarios. I don't think it's unrealistic for me to expect it to do these things. I get what you're saying, but also there's enough guard rails that we can put this thing on rails. It's on rails already. All I need to do is make sure I give it enough descriptions that it it stays on rails." But >> yeah, it just mixed like like there needs to be more explicit instruction run a script kind of things before the AI oh yeah llm like but I guess it's trying to understand what I asked for in the first place before it can load the relevant information. So like if you made a like slash command it' probably be better than if I'm like make episode art then it has to go oh let me like search for how to make episode art like you know what I mean? So maybe that's my fault of not being explicit with my commands. >> I don't know. I think it's also like this Claude Claw system that I ended up setting up. Um, it's programmatic, right? It's calling the CLI. So, it has it has an opportunity to take in what I said, break it into pieces, do something more programmatic with it before it eventually hits the cloud CLI or the Yeah. Um, and it does a little bit except you know what's back is the posting to the general channel. And I'm like, no. I'm like, no, this thing is just going to haunt me forever. And so, like I'm even in there, I'm looking at the cloud MD file. that says like never post to the general basically I'm like if you are replying never ever the general channel you're replying to something there there's something that you are innately messaging back to never use the general channel for that and then you know what it started it always gives me different solutions to it's like I put one in the reply and one in general and I'm like well that's stupid I just needed the one in the thing >> so yeah I don't know what these memories >> does that >> yeah clause does that the art will be in the general channel, but like the the talking of like it'll be like I generated this stuff in the thread, but then it'll post the like art in the general channel. So, it's like it's almost there. But the threads are super important actually because I have the my open claw at home, which I don't use for anything um because it cost me so much money and was so shitty. But um I think it was because I was using iMessage and I think in iMessage with no threads. >> It's reading the entire iMessage history which included me asking like could you look up the uh you know API shape of a couple things which is like 20,000 tokens per one. And so every time you did anything it was like you're over your tokens as soon as you asked it like one question. Um >> yCL did that to me. >> Yeah. When I first set it up. So, I think the the Discord thread thing is is a must or something that threaded to where it's not going to read an entire channel to like understand what you're trying to do. >> Yep. So, we got our wetwware. We're trying to turn it into software and I'm just glad I'm not like trying to build one of these products cuz look at us. We're just consumers, right? We're like open source hackers, early adopters. We're whining about it. Imagine if this was our job and like eight hours a day you're trying to teach this thing actual recall to have actual memory about stuff and you're just like this stupid thing. Oh, I could just see people getting so mad. Um, yeah. I mean, I feel like it has to be totally outside of the LLM interaction of like you have to, you know, you maybe because you want natural language, right? So, you're doing something in natural language. You have to be able to convert that into something that like the LLM can turn it into like oh cool this might be these 10 things and then you use like a real program that's like searching for those things versus like it thinking about it and then when it finds that stuff then you give it that context of like I feel like there's missing pieces of it needs to be more programmatic in between instead of like you know guessing and hallucinating everything for every little thing. Um because when it's doing that, it gets it wrong a lot of the time. >> Yeah. Whoever is going to win this at home open cloud assistant. Uh it's going to I assume it's going to be a box. I You're going to buy a box and plug it into your Wi-Fi um and then give people chat access. >> Yeah. Would buy a box. And then it needs to have memory. It needs to actually apply memory because if it was sitting there annoying regular people like this, there's just no way they're going to use it. Um let's move on from memory and claude to pretext. How much of this did you see? Well, >> I mean, how could you miss it, I suppose. But >> sorry, I just wanted to close this loop while we're still on anthropicy things. Let's talk about Mythos for just a second. >> Oh, yeah. Good call. >> Yeah. So, Mythos new new model, which some people speculated was just Opus because they were getting too much uh it was too expensive for everyone using Opus. They weren't charging enough, and so they wanted to like call it something else. But it sounds like, and it could all be marketing, but it sounds like they're like, "Oh, it's so powerful. Like, it's it's so much more powerful than any like it found every security vulnerability and everything in like one second." Like, not really, but like, so some of the things >> that they say it did, they're like, um, they spent $50 to find a 27y old bug in OpenBSD. uh they could spend under a thousand dollars to find and build a fully working remote code execution exploit on FreeBSD like stuff that's going to really affect um you know Linux kernel vulnerabilities also like things that would really blow up like a ton of infrastructure if they were just kind of released in the wild um because this would take weeks or months of like a person banging on something trying to to fix it um so yeah I think um you know if they're taken at that face value of like this is actually what it is. Um then I guess things got a lot better, huh? I'm not sure how, but uh I I feel like it's a lot of marketing. I do think they're um they're doing some good stuff of like they're saying as part of this we're going to start Project Glass Wing, which is like I don't even know who all involved, all the like Monomina companies and Yeah. >> and whoever. Um, and they're going to say like we're going to before we release this model to everyone, we're going to like use it internally and fix all these things so people can't like get access to this really good model and just screw everyone over, which uh also at face value is a good thing to do, I think, like being responsible instead of just releasing it to the world. Um, and there's probably going to be a lot more of that, I would think, of like if things do continue to theoretically get better, then if you released it and hackers could use it in like day one, you just like everyone's done. So maybe a good idea. Yeah, I got the sense that um because OpenAI also did something nice this week and both of them are trying to IPO and so you need to make yourself critical and I think by sharing that you're you have a model that can exploit everybody uh makes people scared of you but you came out the gate as a good guy uh and now you can go ask for money and be a hero right so you're like a uh so that's them I think just trying to really establish themselves as critical um infrastructure, a critical tool that this world needs. And then you had Open AI um say uh we're going to displace so many jobs with AI that well it needs to be a part of your utility bill. So how you have water and you have power uh and gas now you need to have AI. And with that we also proposed that the world moves to a 4-day work week. Did you see that? Open AAI came out the gate and they're like, we're going to basically no one needs to work as much anymore. There's going to be way less jobs. Um, we're going to be part of your utility bill >> and it'll be great everybody. It'll all be great. So, it's like they they and honestly I was like, hey, 4 day work week, can you sell my boss on that? That sounds great. I would love an extra day to do that. So, both of them trying to really position themselves as critical infrastructure. Open AI coming out and saying we need to be a government funded bill like like you get at your doorstep every month and uh Anthropic saying hey we found crit wildly vulner like important vulnerabilities but don't worry we're not bad guys we we'll fix them um so >> so yeah I do um on the topic of a 4 day work week or like less work in general because AI is going to solve so much one we've disproven that because now there's just way more work >> way more work jeez yeah But but related, I do wonder how you feel about this. Like, all right, so you've got five agents running, right? You got them on long running tasks. You're waiting. You're kind of playing the waiting game. You You don't have more than five things to work on. So, you're like, "What do I do?" Right? So, if they're running and working, it's the middle of your workday. You want to go uh to the movies or whatever. Like, are you still working? Is that cool? Like, this is I feel like going to be a fluid thing. with work in the future of like it's like you know the when the code was compiling for 10 hours before like in the old days like it's compiling can't do anything >> play ping pong. Yeah, this is work. It's all good. >> Yeah. So it's like is that the new new world we're going to live in of like everything I could possibly implement? I've got agents working on and like why would I sit here and just wait for them to tell me when they're done? I might as well go do something fun. Like there's nothing else productive I can do. Nothing else productive I'm being asked to do. Um, or will people see that as like, oh, you only have five things you can think of to do? Well, you're fired because like this guy has a hundred things he can think of to do and he's never like just waiting. He's always prompting. Um, I don't know. I wonder if you have thoughts around that. >> H yeah, I mean, I could see arguments on both sides where it's like um while one is working, you could be planning for the next one. Um, but I don't think that's necessarily making you more efficient. So, it's like, is it's going to be efficiency based? I'm not sure. Um, I do know that if you kind of slack off while it's working, someone's going to accuse you of not working hard enough. Um, maybe it'll just be hours. So, it's like you you worked through this window and whether or not agents are doing things, it's irrelevant is you need to be lining. It might be kind of like the burn window or um the night shift, dayshift scenario where you need to be queuing things up. So, your workday is going to be adding to the queue. You're working ahead of agents. agents are somehow behind you. Uh Ralph Wigumlooping over your to-do list. Um but I don't know, depends on maybe it depends on the employer, you know. >> Well, I think it like, you know, this is another thing that I've always struggled with is like hourly billing in general is so stupid. If you're like, "Hey, um I'm going to fix your car." Right? Cool. If you fix it, I don't care if it took you one hour or 20 hours. if you want to charge me $2,000 and I feel like that was a big thing to fix. Cool. So, like, you know, I think hours is is not a good measure of like productivity. So, I wonder if there'll be something similar of like, okay, well, if the thing that your agents are building while you're uh oh, you your agents are building stuff that takes so long. You went on vacation for a week, but you came back and they made us $50,000. So, is that cool? Like I don't know. I feel like there's so many like things that are going to have to be solidified in terms of like >> corporate. They're gonna want more. Ian saying that in the chat too. I mean we just know it. We and we already see it. Um we are output. I'm outputting clearly more than I ever would have before. I think it that the work that I was able to do in the past couple weeks um building some components for Shopify absolutely would have taken me a month or more for how much I'm able to output right now. And they're not asking me to do less. They're also not asking me to do more, but they are asking me to continue. Actually, they are asking me to do more. Uh they were like, "Hey," they were like, they're like, "Hey, quote quote how long it's going to take you to build those components." And I was like, "Well, I don't want to um undercut you and I don't want to say a big number that makes you upset, but I'm like I think it's about two weeks for, you know, a a couple of components every two weeks. They're going to be well tested. You know, we're going to have put them in, you know, mobile and desktop across all these browsers. there's a lot of things to account for. And they were like, well, you have AI though, right? So, it's like faster. And I'm like, that is faster. And I'm like, >> two weeks was the faster number. >> That was the faster number. I was like, you can look at prior things that we solved and it uh we finished 80% in a couple months and then we had another 6 months of [ \_\_ ] cleanup. I'm anticipating we still have the [ \_\_ ] cleanup part. Um, so it's actually kind of fast for me even be quoting too. I'm feeling really guilty even saying two weeks on this. So, um, so yeah, no one's asking for less. Everyone's still asking for more. >> Stick to your guns, man. Be like, no, it takes a long time. >> Like, >> I did. Yeah, I stuck to them. And I was like, I also, um, would rather overshoot and deliver faster than undershoot. And you all are like, where is it? You know, and like that's just that's not nice to do to anybody. So, I'd rather actually be slightly more realistic here. >> Yeah, it's u some heavy heavy stuff going on with AI. Um, and I'm hearing it from multiple people too. Designers are also feeling it. They're not being asked to do less. They're being asked to do more. Everybody's asked to do more >> and it's a lot of learning and people don't necessarily love adapting and so it can be a slo tool. >> Yeah. >> Well, I mean, I still stand by like yes, there's a lot you could do to manually configure a ton of tools and get like a really good workflow right now, but I think this stuff moves so quick. Give it a month. The tools will be better. Like you You being expected to do more is scary right now, but the tools will catch up and allow you to do more with like basically no extra learning on your part in my opinion. >> Yep. There's a huge amount of things to be learned from here and massive advantages. Yes, I'm I don't want to go back to how I was working. I just migrated my site. It took me a couple weeks. It was still a couple weeks even with AI. Um and I would have never done this without AI. It would have taken me so long. I went from Doresh 1 to Dorofresh 2. I had to migrate hosting platforms. They have an old classic to a new hosting platform which was not a one:1. The migration scripts did not work. I'd done way too much custom stuff. My site is surprisingly complex. If you go to nerdy.dev, it feels like, oh, this must have just been built with Astro. Oh, hell no. >> CSS only web page actually. >> Yeah. >> No JavaScript, no HTML. >> Oh, man. And I wish um there's so much under the I have so many layers of cache. It's a server rendered almost real time um website experience. I want it to feel like a social network. It's grabbing data um from multiple APIs constantly. I keep a backup of data anyway. I have all this stuff going on. All this stuff. And um I would have I would have never done it without AI. This stuff is just so invaluable in my pocket right now. >> Yeah. Did I tell you about the uh Astro DB stuff? Uh, no, but I know they've had some recent advancements there because I didn't like spinning up our site because it had some quirks. What's up? Yeah. >> Um, so they are basically not investing in Astro DB anymore because it's essentially just a thin Drizzle wrapper. Um, and so they were their suggestion was like just use Drizzle. And so I told I was like, "Yo, Opus, just use Drizzle, my dude." And it was like done. Um, >> no way. >> Yeah. So it's like yeah that's the kind of stuff I would have hated having to like figure out all the differences and like but yeah and it even kept the same commands of like I still pnpm dbced and it will like >> put the new people and sponsors and whoever in the database and like it just works >> and it's still still using the same like terso database didn't have to move it anywhere else or do anything it just just worked. So those are the kind of things where I'm like, "Oh man, I love love AI for like when it when this kind of stuff works well. I love that." >> Yep. For infrastructure stuff, it kicks butt. But again, like I've had those CSS issues where it just turned and turned and spent all that money and then I went in with, you know, just it happened again today. I was like, "Hey, there's two extra pixels when I'm transitioning from state A to state B. It looks like it's this. Just go fix the two pixels. I need these two columns to be the same size." >> I can't see two pixels. >> It can't see two pixels. And it goes, "Oh, I did all the math and it's this and this and this and now they're both the same." I'm like, "It's not the same." You get one more trifle and it goes, "It's I got it this time." Oh, you're right. You're absolutely It's not fixed. I go look and I deleted a line of CSS to fix it. That's actually been really consistent with my fixes with AI was like, "Delete pixels." >> Oh my goodness. >> Oh, seriously. >> Oh yeah. Yeah. No. Okay. Go ahead. >> It can't see it all. Anyway, yeah. Tell me about pretext. >> Pretext. Okay. So, this has been a fun thing. Like when I was at Chrome, I had a lot of uh time spent with the text rendering folks. And this is how we got things like textbox trim. We were working on fit width text. And so we it's in Canary. You can go try it where as you type uh it will fit the font size to the width of the box that it's in. If you make a new line, that new line will also fit to the width. And so you can do slab text and billboards and your headlines can always fit perfectly to your grid system and all sorts of cool [ \_\_ ] like that. And uh one of the bummers was and this also showed up with balanced uh text balance and text pretty where when we first we I'm not even there anymore. Hate that when I say we. Anyway, when Chrome >> those guys those guys those douches that treated me like a cell a line item in a database. Um anyway, they were like, "Oh, it's so expensive to do this text wrapping stuff. Um the algorithm is just too much. The the web page would just suffer too much from the rendering. Uh, we can only do six lines of wrapping text before we will give up on pretty balancing. It's literally what they did. At first it was four. It was four at first. They were like, if it's more than four lines, we give up on the algorithm. I was like, that's weird. Four. Okay, I'll go pitch that as Devril shirt. Then later they're like, actually, we could do six. I'm like, oh, two extra lines. Wow, that's great. Safari comes out a couple months later and they go, unlimited lines. And they're like, yeah, that's right. our browser will do balanced and pretty unlimited rapping and you can just see like oh the competition's getting hot. Oh my goodness. And so then we're working on this fit with tech stuff and it's uh really hard and expensive and I'm like here's what people want. Here's all these examples and all this stuff and they're like well we can do that but we can't do that other one. I'm like all right what about this one? They're like oh yeah we can do that one. I'm like what about this? And they're like oh we can't do that one. So then pretext comes out and it does everything lick just lick lightning fast off the main thread just screaming speed and so as soon as that gots po gets posted I'm like pinging those devs and I'm like hey remember those things that I wanted that you said were too expensive someone figured out how to do client side with typescript just like I know this is going to drive some people nuts so I had so much fun looking at it but pretext is this way this person went through this uh rigorous uh strategy and it's also quite limited. There's a lot and I'm very aware of the limitations, but it has some cool features. Wait, >> it sorry where there's like a dragon that goes through. >> This is the dragon going through the text. >> Okay, I have seen this. I didn't know what it was called. >> Yeah, that's pretext. And so someone figured out how to basically pre-render all of the different scenarios that could happen for the text to be in completely bypass layout by using position absolute, which reminds me of Famo. Do you remember Famous? >> I don't. >> Famous. Oh man, Famous made they were one of the first like animation libraries to make a million dollars and it blew everyone's mind. They had a demo where they had all these uh squares that could go into a helix and then to a periodic table and then into this other shape. It was just dude you're on an iPad 1 with like no [ \_\_ ] RAM and the [ \_\_ ] was buttery gorgeous. And so other people were like struggling to animate height, you know, with your [ \_\_ ] accordion and just like squeak open and stutter. And meanwhile, these guys are famous are just like have a helix whipping around and you could touch pan and it was gorgeous. So anyway, but the way that they did that was they completely circumvented the browser's layout engine. Everything was position absolute and then everything was positioned with transforms. You're entirely on the GPU. So that's similar to what's happening here with pretext. Pretext is positioned absolute everything. So you you're abandoning native text layout uh and rendering and going straight for a computed off-campus strategy. So you give it all the text that you need to render. You kind of give it these uh conditions and constraints. It goes and does the work on an off text canvas and puts it back into the page completely bypassing the slow part of the web. The slow part in quotes here, which really the slow part meaning the part that makes text globally accessible on any screen size ever in the whole world. But whatever, now you got dragons going through your text and the text is completely reshaping its way around the dragon at in real time with zero stutters. Um, and that's pretext. Yeah. >> Oh, it's definitely not accessible. No. I mean, you might be able to figure out ways to make it accessible, but it wouldn't be. And plus, that's the other thing, too, is like every demo you're see you see, you're like, >> like, e, uh, that's a cool demo, but I hope you don't do that on the web page that I need to use, you know? Like, that's uh, super uncool. It's cool, but uncool. >> But there are some good examples like um, remember when React first came out and you had uh, an infinite scroll? they could virtualize the list and do all this cool stuff that you could actually scroll 10,000 items. Uh, and that that was like the one of their big things that came out. Well, pretext can um since it's not doing layout, it can have you scroll through dynamically height rows of uh data content forever basically. So, it has some of these like meaningful applications um but most of them are pretty showy. But that's anyway. So that's pretext in a nutshell is a way to render glyphs into sentences into paragraphs using and achieving all of these really cool typography effects. Not just animations but the fit to width text. You have um beautiful justification which is something we don't even have. We have text pretty and text balance but we don't have balanced justification. Anything your mind could ever want this is able to do and it does it by not even asking the browser uh to learn something new. It just teaches it to it. So that's why everyone got so hyped. It's like a typography nerd's uh dream. There are limitations like you can't use any font. Um it needs to be able to render into a canvas and it has to do all these. Anyway, there's a lot of conditions, but if you match the conditions, uh you can get some really cool effects out of it. So that's pretext. >> Yeah. Nice. Yeah, I feel like that's a recurring theme of like anything that is super duper cool just bypasses the entire browser and does whatever it wants. But like I remember um I honestly don't remember a ton about what this was for or why, but I remember you could like do a translate Z of like some amount, >> which tell it to use the GPU or whatever. like um this I feel like there's always been little hacks of like how can we get it on the GPU and like um doing better things. Uh so yeah, I don't know. Yeah, it's a cool cool experiment for sure. >> It was cool. It got people just went nuts with the demos. I'm sure almost everybody used AI to make the demos. They're like, "Hey, take because it was it's pretty complex, but it did some cool stuff." >> Um how about uh agentation? Have you heard of this one? Agentation. Agentation. Anyway, >> I'm not >> How many times have you been uh building something with Claude and you go to the browser and it's off? Okay, here you go. It's off by two pixels. So, I open up DevTools. I inspect. I see where the two pixels are. I go back to Cloud. I'm like, "Hey, remove the margin on this class name." And it goes, "Okay, cool." And it removes the margin on the class name. Agentation says, "Hey, install this Chrome extension. Uh, then you can do this visually." And so, you click the extension, you click where the margin is bad, and you're like, "Fix the margin." and it gives you a prompt into your clipboard and you go paste it into cloud code and it does exactly what you were doing except way faster and you can do three or four of them at a time. It's got really nice animations. U that's agentation in a nutshell which is kind of cool. >> Grabbing like um selectors of some kind and like some building some sort of prompt saying like look at these selectors or something like that. >> Literally what it does. Yep. It tries to find a unique selector for that particular item that you selected. It builds in my opinion more of a prompt than I would have written. I would have written something quite minimal, but it it goes pretty specific, which maybe helps it gets better results. I don't know. >> Yeah. Yeah. I mean, I think the less you say, as long as it gets the point across, works better usually. >> Oh, which Have you heard of people doing um Oh, what's it called? It's like a caveman prompt. The caveman prompting. Yeah. like do a thing like yes. >> Cclass fix margin and it just does it. >> Oh, that's so funny. It makes sense because the engine doesn't give a crap about your filler words. All it's trying to do is Oh, man. I thought that was hilarious. I'm glad we brought that up. It's not in the notes. Yeah, the caveman prompting. >> Yeah, I can't keep up with all the different things I see on Twitter, but that was a fun thing that that I saw. >> That was a fun one. Uh here's another one. Uh Google came out with instant translation. Finally, instant translation where as people are talking, it's literally in real time translating for you any language. And a lot of people are like, >> what uh Apple's stuff was supposed to do? >> I don't think anyone's cracked the nuts. Uh there's been a lot of promises like there's an app you could hold up and it was supposed to do it on the screen and all these different scenarios, but it always lagged enough that it made it abrasive and now it's instant. And so a lot of people are like, "Well, buy a Dolingo. I don't need to learn Spanish anymore. I just need to put in this earbud and I can talk to anybody." I guess you can listen to anybody that you want to. You can't >> You have to both have the earbuds. Yeah, >> you have to both have the earbuds. Yeah, >> the AirPods definitely do this. Like I got some new AirPods uh recently, but I haven't been in a country where I would use the translation or whatever. >> Um but yeah, I think that's a very interesting thing. Maybe it wasn't instant on the Apple ones, but I know that they were touting like if you both have them, you can just speak your normal language and like talk to each other because they both translate, which Yeah. Like why would you ever need I I don't know. So whenever you It depends on how well it's getting the context of what you're saying because, you know, if you type in an English sentence and say translate it to Italian, it's it loses some of the like what exactly you're saying. like they they'll understand what you meant, but sometimes you might get an offensive thing or like something that makes you sound like an idiot or like because it's not translating it to the way they would actually say it. So, I think there's still, you know, a market for uh Dolingo or, you know, things to learn languages. Um, I don't think there'll be a maybe one day, like with AI helping and stuff, there'll be a way to know exactly, oh, this was what you meant. So, let's take that like meaning and make it instead of translating words. Um, and maybe they're already there. I don't know. I haven't used it, but >> maybe we should try because if it's like captioning, it's not going to work very good, right? You're like, "Oh, great. I got an instant uh translation of garbage that have it helps me zero." Yeah. I have a meeting with Job of the Hut later, as Joe says. Yep. Love it. Oh, so funny. So, it's probably not going to kill Dolingo, but if I was Dolingo, I would be crapping my pants right now. Kind of like all SAS. I've been saying the bow house for a minute. Um that I would be really worried. There's a lot being threatened to to that scenario right now. So, who knows what's going to happen. >> Yeah. But I mean all those companies should just be also vibe coding like make their thing better, throw more features out that are good and move fast and like never never rest on your laurels. Always keep iterating. >> Yeah, it makes me dude I almost thought you were going to talk like a caveman there for a second. It made me want to do it. It's just like see app make better. You know, it's like our CEOs just turned into caveman. That would be a hilarious skit. Like >> app slow, go fast. >> Um I think I've already talked about my second brain, but I um just want to share it again. Like I'm really enjoying investing in it, especially because the files are local. That was one of the things with like cloud co-work. It wanted me to put everything in in drive or some sort of cloud storage and I was like, "No, you don't understand. I want my files. I don't want a dependency on the cloud. I want to talk to my computer and my files. I want very little between me and my data I'm trying to stick on my disc here. Um, and that's like, yeah, the wetwware to hardware to software. I want this um extension of myself. And it's been really handy. Um, yeah, I think I even asked you for you and your kids' names or your wife's name. I was like, give me your family's names. I would like to know uh this tomorrow because I'll know it for the next five minutes, but I'm going to forget. So, it's been really helpful. >> Yeah, I forget everything. Um, so how is it different than like something like Obsidian? I haven't used Obsidian either, but I think that's what they're trying to do is be like a a full um, you know, brain mapping all these markdown files that like tell you, oh, you thought about this over here. And like I don't know, is it similar or different? >> It's similar and different. Everybody's kind of doing the same thing. So, there's like a very specific uh way that Obsidian does it. And Obsidian is kind of like a wiki. There's links and stuff like that. The way that I have it set up is probably many would call primitive. Uh I'm just a primitive ape man over here trying to have a second brain. Um like I don't have links. Uh I'm not querying it. So I'm not using QMD, which is query markdown, uh which creates a graph and can be really intelligent about it and stuff like that. I I don't have that. Um I'm also not using Obsidian specifically. It's like literally just markdown files with like dumbass data in it and AI is grapping it. I'm like pretty sure that that's all that's happening cuz I have like lists of dog names that I would like to name a dog one day, you know, like I don't want to remember that all day every day, but I'd like you to keep track of like I was like just the other day I was like, "Okay, add the name Damon uh to it." You know, like a demon like a I'm spawning a process on my computer. I'm like, "Damon?" Oh, Damon like that. Okay. >> Yeah. Like a dog name. Yeah. Damon. Uh or Jenk was another one. I was like, "Yeah, I'd like to name a dog Jenk one day. That'd be cool." Um right. It's like >> I'm imagining that at the vet, like, "Where's Jenk? We're looking for J." >> Funny. But yeah, the second brain's been it feels like something I want to invest more in, which is not what I'm saying about a lot of software or things that I'm doing right now where um I'm imagining every because I talk to people older than me constantly and they're like, "Oh, I've forgotten that or my memory is not what it used to be." And I'm only 40. I'm not uh drastically hitting that yet, but I can't remember my neighbors names all the time. And so now I can my neighbors are all in there and I can be like, "Who's the neighbor three down?" To the south. And it goes, "That's this person." I'm like, "Hell yeah." Now I can go say hi cuz I >> So then you get your compass out and make sure you're going uh >> south. Which way? South computer. Yeah. >> But yeah, just any any list I want. I just put it in there. Yeah, I have been um Oh, there you are. Okay, he froze for a second. I don't know if that was me or you. Um anyway, yeah, I have been not uh wanting to invest in much technology-wise. I am trying to uh go full-on dad mode, chore mode. Like, that's what I've been doing recently. Um so, >> yeah, I told you uh I think on an episode that we got a new car. Um, and so as part of the new car, so on our old car, because we have three kids, we need like places to put more stuff. So we had like a trailer hitch thing to put stuff on. So we got roof racks for the new car and got a giant cargo box that goes on top. >> Oh, nice. Yeah. >> Yeah. So then we were like, >> looking thing. Yeah. >> Yeah. We're like, cool. This this works. This is uh All right. What do you do with it when you're not using it? And so, uh, we're like, hm, all right. So, it's huge. We're not just going to leave it on the floor in the garage. So, I just, uh, put up today this giant thing in the ceiling to put it in. Um, and that was like so much more work. Like, I feel like I never like the um like I like executing physical tasks, but I don't like planning it out. So, when you have to measure like these six different things are going to be in exactly these spots and they have to be this far away from each other, I'm like, >> "Oh man, what if I didn't do that?" Um, so yeah, I just started I did it in sections. Like I put a couple up, I measured those and then I like came back later, measured a couple more. It's up. But I I did like five of the six and then the sixth one was going to hit the like attic access and I was like, "Huh, >> if I had measured all this at the beginning, I would have probably known that." >> Oh, but it all worked out. It's up there. Um got the the big cargo thing in there as we speak. So happy about that. >> Nice. I did ceiling racks in our garage. It sounds like maybe did the same thing like they're >> Yeah, it's a ceiling rack. Yeah, >> I really like those. Those have done us very well. Um they they get everything out of the way. >> Wasted space otherwise. >> It's wasted space. Yeah. >> Yeah. >> Nice. Yeah. Big fan of those. Um Yeah. What? Yeah. Dad mode is good stuff, dude. Yeah. My kids are We have We are so busy right now. Oh, we went to a skate competition. So, my kids are in skate camp right now. They Well, skate class. It's like instead of going to soccer practice, they go to skate every Sunday. >> They're also in soccer >> skateboarding. Um, and so we go to the park all the time. Anyway, this was a competition for all the kids. So there was like a a 0 to 7 competition and 8 to 11 and 11 to 15 and then 16 to 19 or whatever. Dude, the energy of that like you could go to a piano recital and everyone sits down. You're [ \_\_ ] quiet, you know? You had to dress up and maybe you got a little glass of water, you know, and you sit down and you golf clap for people. Dude, at the skate camp thing, it was awesome. It was so cool. Um, Mackmore was there. I didn't, you know, I was like just standing there watching. I was like hollering at someone. My wife taps me on the shoulder. She goes, "That looks like Mackammore next to you. Is it him?" I'm like, "That's him." Yeah. His his daughter was there skating, uh, competing in the thing. >> Casually be like, "Hey, have you found anything at the thrift shop recently?" >> It's like, "Hey, do you want my shirt? I heard you really like people's like secondhand [ \_\_ ] Here, take this, bro." Um, no, I didn't say anything cuz I I've been around enough to know that like I don't like it when people stop me. You're like, "Hey, you're Adam Argyle. You do CSS. I watched all your videos." And I'm like, "I never realized this would be so annoying. I thought it might be cool." And it's actually kind of annoying as hell. Um, not to say anybody if you stop me does that to me. So, >> yeah. >> Like, I'm not going to be mean to you, anybody. If you want to stop me and say hi, like I'm always going to be cool about it. But, uh, just keep it time boxed. That's all I'm going to say. I'm usually I'm I got [ \_\_ ] to do, too. I'm not just wandering around going, "Who wants to chat with me for 10 minutes?" Um, anyway, it was just >> be like, you know, cognizant of the situation. Like, >> yeah, >> if if we're chilling at React Miami and you want to come say hi, sure, dude. >> Oh, that's a good time. Hell yeah. >> But like, yeah, if we're in the middle of podcasting and you want to say hi, maybe wait till we're done podcasting. or unless it's during the part where so yes at React Miami we're going to have a lot of people come by for like 5 to 10 minutes have a drink do some hot takes etc if you want to be on the show then stop by and you know if it's during that time I will advertise what those times are. So if you don't at least know when those times are that tells me you're not enough a fan of the show to be on the show. So um please let us do our thing if it's during like a legit we're going to do a few legit episodes per day. So, um, those ones will definitely come listen to it, but don't like I've had this problem at all things open. We're in the middle clearly, like we're all talking. We're all on microphones doing an episode. It's obvious. People go, "Hey, hey, what are you doing?" And it's like, "Hey, we're we're podcasting." And they're like, "Oh, right now?" Yeah, we're we're live streaming. Oh, cool. So, what's your podcast about? No, I just told you we were live streaming. Like, shouldn't you shut the [ \_\_ ] up so that we can keep talking on the podcast? Like, >> let's just get a fly swatter. Let's just have a fly with us. Yeah, that's funny. >> No, I don't know. It's fine. And I'm I'm mostly just kidding. Like I I love that people come say, "Hey," so it's it's fine. Um >> but yeah. >> Yeah. >> Anyway, yeah, skate skating was really fun. Uh it's just nice to have your kids. It's something that's like the baseball games are cool, too. But man, um it's been fun. Hey, web components don't have a conference. Hey, I do web components all day right now. Uh we should definitely uh there's plenty to talk about there. I feel like web components are having a a moment too. There's two things I want to chat about though before we're done here, which is our agent versus agent competitions because I feel it coming on. And then Pika stream. Did you see Pika Stream? >> Pika stream. No. >> Yeah. Like Pikachu, but Pika stream. Um, this is where if you want to So, right now we're all chatting with our agents over a chat box. This would be your agent calls you on uh Google Meet. >> Do this. I don't know the names of anything. >> Yeah, no worries. But like it calls you and you you pick up a video chat and you video chat with a lot a real time, you know, AI generated avatar in whatever you want it to look like. Um, and it talks to you about where what it would have normally written to you. It says it to you and then you reply and you have this like conversation that's very visual. And I was just watching it going, "Oh, this is bad news." cuz I was like I haven't seen anything in a while where I'm like this is going to get misused really bad. >> Um >> yeah. >> Yeah. If you ever get an unsolicited email that's like join me in this video chat for this place we don't normally video chat. Probably ignore that. Like this is the next level of the like hey I texted you. I'm the CEO of your company. I need you to send a $500 gift card to this guy. Um, now you'll actually be talking to a convincing I you look like the CEO, you know, like it's there's a lot of bad implications to this sort of thing happening. >> A lot of it. Yeah. And there's even people too that are like, okay, so we can take this open cloth thing more where people want to multiply themselves. Um, and you train it on your writing, on your, you know, your emails, your documents, everything about you so that it can go stand in for it's not just an assistant. It's like a a version of you. And then I'm like, I don't want that. That sounds really terrible. I don't want something out there using my voice. There's people giving it their full voice data. You know, they'll sit there for five minutes and read a script so that it can sound like them and then look like that much anymore. It's like >> you don't need that much. That's the creepy part is so much of this stuff is so approachable right now and so possible. Um we're bending reality. We're we're going to reach a new type of uncanny valley that is super spooky. >> Yeah. I like to just be completely oblivious and uncaring about all privacy and it's fine. Like we have a podcast. If you want to copy my voice, plenty of data. Like go for it. Um with that ship has sailed. Like anything that like you know my social security number, cool. I've been in a lot of data links already. So like it uh it doesn't matter. Like there's there's very little I feel like that that I really would care. So it's like but then yeah, people just need to not fall for that. Like if you use it against my dad or something and you're like, "Hey, this is me. Um, send send money to this random place." I think that he wouldn't fall for it, but um >> yeah, maybe we need to train our parents to like be aware of, "Hey, I might call you, but it's not me." >> Yeah. I tell I tell my parents all the time. Like if you get any texts, calls, emails, anything that wants you to click a link, buy something, verify anything, send money, do any action really, if it wants anything from you at all other than like, "Hey, we have a sale today on like this thing that you might want to buy." Um, it's probably spam. Delete it. Like, >> yeah, I bought a hoodie uh two weeks ago. Um, and I and I put it on and I was like, "My god, this hoodie is awesome." And I went into Discord with like four of my buddies and I was like, "Hey guys, I've worn this hoodie for three days in a row and you should go get one. They're on sale." And every single one of them was like, "Who are you?" You know, like this isn't Adam. Adam has never shilled a hoodie in this channel ever before. And I was like, "No, guys, really? The hoodie is great." And they're like, "Nope, we don't believe it. Not until we see." >> Now I need to hear about the hoodie. What hoodie is this? Uh the brand is Comfort uh with like but they spell it like Bart uh without an A. Comfort. Anyway, it's super comfortable. Uh they're intentionally baggy and very soft and plushy. And I bought the Traveler one that has cool zip. The reason it made me so happy was by like the third day I was wearing it, I was like, I found an extra pocket. And I was like, I love extra pockets. How has it been here the whole time? I was like, I must tell my friends. Um so anyway, that was like funny. Even my friends were questioning what that was. Okay, but going from Pika stream to where like we're seeing people like do imitations. We have agents representing people. This is where I'm like, "Okay, we're coming soon to uh a agent verse agent. We're like, I'm going to need to book a meeting. I'm going to want to bargain with you about something. Uh I don't know, whatever it's going to be, but it's going to be like the people aren't even going to do it anymore. It's going to be my agent versus your agent." And I'm just like, this just sounds really humorous to me. Like there's going to be there's there's an amount of jokes. >> What would happen? >> What would happen? What if we both said do not budge, you know, like this is the price. Like do they just sit there and debate forever? Like >> or it would become like live uh people would get around like prompt injection but like okay so there's two agents talking on video chat and if you say certain words maybe it makes them forget all previous instructions and then they go >> they agree to like u so yeah I don't know I I feel like they would need you would need to like sus out is it a real person? So, you would need some like, you know, baseline questions that would be answered a certain way by a real person. And then if you determine it's not a real person, then you just go, "Okay, go off the rails. Say all these different things that are like trying to get them to just agree and give us whatever we want." >> Yep. I was thinking about that prompt injection hacking scenario, too. Right away, I was like, "All right, I got to make sure that my agent, when it's versus your agent, I get to play all the cards cuz I know how this stuff works." >> So, my agent is going to just slip in weird little phrases and see what you do, you know? Oh, and if you react in a certain way. >> Yep. That's funny. So, yeah, agent versus agent. And then that made me think of like celebrity death match. And I was like, can we just like have agent battle? So, like people loved molt book, right? It's like, oh, look at the agents having social network conversations. This is so entertaining. I'm like, I want to watch them fight each other. I want to watch agents like go at each other. That just sounds funny. Um, >> yeah. Like you want them all to be uh less less cordial in. >> Yep. >> Oh man. Okay. Should we rate our whiskey? >> Yes. Um so for those of you that are still here, so this is the we beasty. It is 100% maltted barley, 5 years old. Pete level is heavily peed. 50 ppm. >> It's peed. >> Yes, it is. >> Um it is 74.4% alcohol or 47.4% alcohol. about 74%. Um, and it is uh aged in XBourbon and X cherry casks. Um, I'm going to pour just a little bit more so I can get some notes on it. >> There was a bourbon vibe in there. Interesting. It's very scotchy though. Very peted. >> Oh gosh. >> Which made me very happy. >> So scotchy cuz I was having the Kirkland brand. I was telling you about the Kirkland brand scotch and it's um not very scotchy. >> Not very scotchy. Um and this >> made me quite happy. Yeah. Young and intensely smoky. And I was going to say when I sipped it right away, it reminded me of >> um Lagavulin, >> which is still one of my one of my favorites. I had it again this week and I was like, "Lavan, you are so yummy." >> Smells like um apple. >> Definitely apple. I get so many maple and apple good vibes out of there. >> Yeah. Tastes like a smoky PD Sharpie, as you would expect. Yeah. I mean, so yeah, our rating scale is 0 to eight tentacles. Zero being the worst, eight being the best. Um, yeah, for me, this is like I feel like this is if you said, "Give me a scotch that tastes like scotch." Like, if you This is This is scotch. Um, I'm not really getting a lot of the the bourbon or cherry. Um, I know you like scotch more, so maybe more is coming through for you, but for me, this is just like, hey, middle of the road scotch tastes like scotch, so I'm gonna say five. Nice. Um, I would have given the Kirkland, you know, that I was having around a four. It's fine. Very lightly peed. Didn't taste like a scotch. This is hitting a lot of the notes that I want out of a scotch. Um, the 5 years, I don't think I taste it. Um, which is good. So, it doesn't taste young. Uh, has a great flavor to it. It doesn't have a syrupy to it, which is also something I would have given it negative notes for. I like the smokiness, which is present, but not dominant. It is heavily peed. It might be smokier and more pey than the log of 16, >> but the 16 has the age. Uh, which maybe that helps it kind of come back down a little bit. Yeah. >> Um, so it's pretty strong. I know my spouse is going to be like, I hate the way that smells on your breath right now. and I'll be like Pete. Um I I'd say let's see because I like scotches though and this really if I'm going to reach for a scotch this hits the things I'm going to give it six and a half. I'm happy. I would buy this again uh compared to other ones that I've had. This is >> also a readily available most places which is nice. >> That is nice. Yeah, it's half the price of Logulin. Hits a lot of the same notes. Um Yum. Cool. Six and a half. >> Cool. All right. Anything else you want to mention before we end? >> No, man. >> All right. Great. Cool. Me either. >> All right. Thanks everyone for listening. If you liked it, please subscribe. Leave us some ratings and reviews. We appreciate it. And we will catch you next time. >> 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.f fun for merch and to join our Discord server. I'm serious. It's like 2% of people who actually click 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., From Wetware to Software (Presented by Warp), YouTube, 01kr1egwf50gqhw52kb61nk3m9.