The Manager Has Become The Managed - CourtneyR.dev's 2nd Brain

The Manager Has Become The Managed ![rw-book-cover](https://i.ytimg.com/vi/Zja5HO4Ztp4/sddefault.jpg?sqp=-oaymwEmCIAFEOAD8quKqQMa8AEB-AH-CYAC0AWKAgwIABABGHIgRSg7MA8=&rs=AOn4CLADoIItXSNTFkCdEGUp6zuLA7jSvA) 🔗 **Source:** [The Manager Has Become The Managed](https://www.youtube.com/watch?v=Zja5HO4Ztp4) > [!note]+ Document Notes > tags: [ai-development, developer-experience, automation, software-engineering] > author: Adam Argyle > source: YouTube > reference: https://www.youtube.com/watch?v=Zja5HO4Ztp4 > date created: 2024-12-19 12:00 > > # 🏷️ Title: The Manager Has Become The Managed >  ## 🧰 Action Inbox - [H1] Explore AI agent orchestration tools for managing code review bottlenecks — addresses role reversal from manager to managed #automation - [H2] Implement spec-driven development workflows to maintain control over AI output — prevents overwhelming PR creation rates #software-engineering - [H3] Evaluate workspace-based development environments that bundle all tools — reduces context switching between terminals/apps #developer-experience ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [Intent - Workspace-based Development Tool](URL) - [Augment Code](URL) - [OpenClaw](URL) ## 📝 Summary - Developers are experiencing a role reversal where AI agents create more work than humans can review, making managers feel "managed" by their tools - AI chatbots may not be the optimal interface for development workflows; specialized tools with embedded context work better - Traditional development craft is evolving from writing code to orchestrating AI agents and systems - The backlog has become valuable again as AI can quickly implement previously stalled ideas - New workspace-based development environments bundle all necessary tools to reduce context switching ## 📖 Main Content The conversation explores how AI is fundamentally changing the developer experience, with a focus on the unexpected reversal where developers find themselves overwhelmed by AI-generated work rather than empowered by it. The speakers discuss how traditional chatbot interfaces are limiting and explore emerging solutions like workspace-based development environments. Key themes include the evolution from individual coding to agent orchestration, the renewed importance of planning and specification, and how the craft of programming is shifting toward higher-level system design and management. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/Zja5HO4Ztp4?t=0) — [H1] Introduction and whiskey tasting setup - Key idea: Setting up discussion about AI development tools with guest Amelia Watenberger - Evidence: "This is Amelia Watenberger, digital creatives, products spelunker, codebase revealer, data visualizer" #### ⏱️ [12:30](https://youtu.be/Zja5HO4Ztp4?t=750) — [H2] AI chatbots as limitations rather than solutions - Key idea: Chatbots are jack-of-all-trades but master of none - Evidence: "chatbots are kind of the terminal of the next phase... they do everything pretty well but they don't do anything really well" #### ⏱️ [18:45](https://youtu.be/Zja5HO4Ztp4?t=1125) — [H1] The manager has become the managed phenomenon - Key idea: AI agents now create more tasks than developers can handle - Evidence: "a bot started creating tasks for me... all of a sudden I am now the other end of the vibe kanban" #### ⏱️ [25:20](https://youtu.be/Zja5HO4Ztp4?t=1520) — [H2] Loss of programming craft and satisfaction - Key idea: Fast AI solutions eliminate the struggle and victory cycle of debugging - Evidence: "I used to run, I used to get up at office all the time... now it happens with AI in like one prompt" #### ⏱️ [32:10](https://youtu.be/Zja5HO4Ztp4?t=1930) — [H3] Backlog renaissance in the AI era - Key idea: Previously dead backlogs are now actionable with AI implementation - Evidence: "Joel Hooks was saying that our backlogs cool again... Your backlog used to be a dead weight" #### ⏱️ [38:50](https://youtu.be/Zja5HO4Ztp4?t=2330) — [H2] Burnout and new types of stress with AI - Key idea: AI doesn't reduce stress but creates different kinds of unhealthy relationships with computers - Evidence: "I'm in a new type of unhealthy relationship with my computer... I'm not better. I'm not doing less" #### ⏱️ [45:15](https://youtu.be/Zja5HO4Ztp4?t=2715) — [H3] Interface similarity problems across AI tools - Key idea: Similar chat interfaces cause confusion between different AI agents - Evidence: "I find myself... I'm not always very polite to my agents... I've almost typed agent things into Slack" #### ⏱️ [52:30](https://youtu.be/Zja5HO4Ztp4?t=3150) — [H1] Voice-to-visual ideation experiments - Key idea: Exploring new mediums beyond text chat for AI interaction - Evidence: "transcribed as you were talking and then turned your concepts into index cards" #### ⏱️ [58:20](https://youtu.be/Zja5HO4Ztp4?t=3500) — [H2] Data liberation and personalized applications - Key idea: Moving from siloed apps to personal data control and custom experiences - Evidence: "I'm liberating my data... when it's mine, I can build my own app with it" #### ⏱️ [65:40](https://youtu.be/Zja5HO4Ztp4?t=3940) — [H1] Intent workspace-based development introduction - Key idea: Bundling all development tools into unified workspaces - Evidence: "if we put everything to do with a task into like a bag... we get like the crazy orchestration workflows" #### ⏱️ [72:15](https://youtu.be/Zja5HO4Ztp4?t=4335) — [H2] Spec-driven development with AI agents - Key idea: Using specifications to control and organize AI-generated code - Evidence: "I write down what I want to do. An agent goes and explores my codebase and writes a spec for me" #### ⏱️ [78:50](https://youtu.be/Zja5HO4Ztp4?t=4730) — [H3] Evolution from IDE-based to workspace-based development - Key idea: Moving beyond traditional IDEs that were designed for manual coding - Evidence: "what if we didn't use IDEs anymore because it's been pretty restrictive trying to design within the IDE" ## 📚 References - [Intent Desktop App](https://intent.dev) - [Augment Code](https://augmentcode.com) - [Fullstack D3 and Data Visualization](https://www.newline.co/fullstack-d3) - [GitHub Next Team](https://githubnext.com) ## 🗣️ Paraphrase - AI development tools are creating a role reversal where developers feel managed by their agents rather than managing them, as bots generate more work than humans can review - Traditional chatbot interfaces are insufficient for complex development workflows, leading to exploration of new interaction paradigms like voice-to-visual ideation and workspace-based environments - The craft of programming is evolving from manual coding to orchestrating AI systems, requiring more upfront planning and specification while potentially eliminating the satisfying struggle-victory cycle of debugging ## 📰 Newsletter The future of developer experience is being reshaped by AI, but not in the ways we expected. Instead of making developers more productive, many are finding themselves overwhelmed by AI-generated code and pull requests they can't review fast enough. This "manager becomes managed" phenomenon highlights a critical need for better orchestration tools and interfaces beyond simple chatbots. New approaches like workspace-based development environments are emerging to bundle all necessary tools and context, while spec-driven development helps maintain human control over AI output. The craft is evolving from writing code to designing systems and managing AI agents. Resources: [Intent Workspace Tool](https://intent.dev), [AI Development Best Practices](https://developer.wordpress.org/), [Agent Orchestration Patterns](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer advocacy: Create content about managing AI agent workflows, present at conferences on "post-chatbot" development experiences - Education: Develop workshops on spec-driven development and AI orchestration patterns - Developer experience: Research workspace-based development tools, evaluate developer productivity metrics with AI - Developer marketing: Document case studies of successful AI-human collaboration workflows - Community management: Survey developers on AI workflow pain points, moderate discussions on tool fatigue #### OSPO - Policy & Governance: Establish guidelines for AI-generated code review and approval processes - License Compliance: Review AI-generated code for license compliance, train teams on provenance tracking - Security Management: Implement security scanning for AI-generated pull requests and automated code changes - Software Inventory: Track AI tools and agents in development workflows, maintain SBOM for AI-generated components - Developer Support: Provide training on AI agent management and orchestration best practices #### Community - Maintainers: Create templates for managing AI-generated contributions, establish review workflows for automated PRs - Contributors: Develop onboarding materials for AI-assisted contribution workflows - Sponsors: Budget for new categories of developer tools focused on AI orchestration and workspace management #### WordPress - Core: Evaluate AI-assisted development workflows for WordPress core contributions - Core Performance: Use AI agents for automated performance testing and optimization suggestions - Core Program Team: Coordinate AI tool adoption across teams, establish standards for AI-generated code - Documentation: Explore AI-assisted documentation generation and maintenance workflows - CLI: Investigate AI agents for WP-CLI command development and testing automation ## 🧪 Hypotheses - Workspace-based development environments will replace traditional IDEs as AI adoption increases and context switching becomes the primary productivity bottleneck - Spec-driven development will become the dominant paradigm as developers need structured ways to control and verify AI-generated code output - The traditional code review process will evolve into AI output validation, requiring new skills and tools for managing automated contributions - Voice and visual interfaces will supplement text-based AI interactions for complex ideation and system design tasks ### 📊 Methodology - Conduct user studies comparing productivity metrics between traditional IDE workflows and workspace-based approaches with AI integration - Analyze code review throughput and quality metrics when managing AI-generated versus human-written code contributions - Survey developers on stress levels and job satisfaction before and after adopting AI agent orchestration tools ### ❓ Questions - How can development teams maintain code quality and security when AI agents generate code faster than humans can review it? - What new interface paradigms beyond chatbots will emerge for complex AI-human collaboration in software development? - How will the role of senior developers evolve from individual contributors to AI system architects and orchestrators? ### 🔮 Future Research - Investigation into natural language specification languages that can effectively control AI agent behavior and output - Development of collaborative AI-human code review systems that can handle high-volume automated contributions - Exploration of multi-modal development interfaces combining voice, visual, and text interaction patterns ### 💡 Implications - Practice: Development workflows must evolve to handle AI-generated code volume, requiring new review processes and quality gates - Policy: Organizations need policies for AI agent behavior, code ownership, and responsibility for automated contributions - Security: New attack vectors emerge when AI agents have broad system access and can generate code autonomously - Education: Developer education must shift toward AI orchestration, system design, and specification writing skills - Funding: Investment priorities should focus on developer experience tools that manage AI complexity rather than just AI capabilities ### 🧠 Learning Styles Assessment Reading level estimate: College/Graduate Level (complex technical concepts, industry jargon) #### 👀 Visual learners - Present: Discussion of visual interfaces, workspace layouts, kanban boards - Improve: Add diagrams showing AI agent workflows, workspace organization patterns, code generation pipelines #### 🔊 Aural learners - Present: Full audio discussion with conversational explanations - Improve: Create structured audio summaries of key concepts, add timestamps for easy navigation #### 🗣️ Verbal learners - Present: Extensive use of analogies (rubber duck debugging, craftsmanship metaphors), defined technical terms - Improve: Add glossary of AI development terms, more real-world examples of workflow changes #### 🧩 Kinesthetic learners - Present: Discussion of hands-on tool usage, workflow experimentation - Improve: Provide step-by-step tutorials for setting up AI development workflows, interactive demos #### 🧠 Logical learners - Present: Clear cause-and-effect relationships between AI adoption and workflow changes - Improve: Add decision trees for choosing AI tools, workflow optimization frameworks #### 🧑‍🤝‍🧑 Social learners - Present: Collaborative discussion format, team workflow considerations - Improve: Add discussion prompts for development teams, peer review frameworks for AI workflows #### 🧘‍♀️ Solitary learners - Present: Individual developer experience focus, personal workflow optimization - Improve: Self-assessment tools for AI readiness, personal productivity tracking methods #### 🌿 Naturalistic learners - Present: Evolution metaphors for development practice changes - Improve: Ecosystem thinking about tool interactions, sustainability considerations for AI resource usage ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear heading structure, logical content flow - Improve: Add ARIA landmarks for major sections, ensure all technical concepts have descriptive context #### Keyboard navigation - Present: Standard navigation patterns for video content - Improve: Provide keyboard shortcuts for timestamp navigation, ensure all interactive elements are reachable #### Contrast and readability - Present: Standard video player interface - Improve: High contrast mode for video player, adjustable text size for any overlay content #### Text resizing and reflow - Present: Responsive transcript layout - Improve: Ensure content remains usable at 200% zoom, test with various viewport sizes #### Media and captions - Present: Full transcript available - Improve: Add closed captions with technical term definitions, provide downloadable transcript #### Forms and interactions - Present: Basic video controls - Improve: Clear labels for all interactive elements, accessible error handling for playback issues ### 🌍 DEIB Assessment #### Cultural breadth - Present: Discussion focused on general developer experience across different contexts - Improve: Include perspectives from developers in different geographic regions, varying company sizes #### Inclusive language - Present: Gender-neutral technical terms, explained jargon - Improve: Avoid assumptions about team structures, include diverse workflow examples #### Neurodiversity support - Present: Clear topic transitions, consistent structure - Improve: Provide content warnings for discussion of work stress and pressure, break long sections into smaller chunks #### Representation - Present: Female guest expert, diverse technical backgrounds - Improve: Include perspectives from developers at different career stages, various technical specializations #### Psychological safety - Present: Respectful discussion of tool limitations and developer frustrations - Improve: Acknowledge that not all developers have access to cutting-edge AI tools, validate concerns about job displacement ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Clear progression from AI capabilities to workflow implications to tool requirements - Inductive strength: Good use of personal experience examples to support broader trends - Causal reasoning: Well-established connections between AI adoption and developer experience changes - Analogical strength: Effective use of management analogies and craft metaphors #### Fallacies - Appeal to novelty: Some assumption that newer AI tools are automatically better without considering all trade-offs - Hasty generalization: Limited discussion of developers who may prefer traditional workflows or have different needs #### Tone of voice - Overall tone: Conversational and exploratory, balancing enthusiasm with realistic concerns about AI development tools ### 🛡️ DARVO Analysis - Deny: No relevant instances - Attack: No relevant instances - Reverse victim/offender: No relevant instances - Emotional cues and rhetoric: Positive and collaborative discussion focused on problem- --- ## Full Transcript [00:01] Adam Thomas Argyle the nerd. Howdy [00:05] y'all. [00:07] What's up? What's up? All right. We are [00:09] joined with a special guest today. You [00:11] want to tell the folks at home who we've [00:12] got, Adam? [00:13] >> Oh, you know I do. This is Amelia [00:15] Watenberger, digital creatives, products [00:17] spelunker, codebase revealer, data [00:19] visualizer, mental modeler, badass dev, [00:21] previously principal research engineer [00:22] at GitHub, known for code atlas, [00:24] co-pilot for docs, GitHub blocks, and so [00:27] many more. It's currently at Augment [00:29] Code, author of Fullst Stack D3 and data [00:31] visualization, and overall quality UIUX [00:33] design engineer. Welcome to the show, [00:35] Amelia. [00:36] >> Hey. Hey. Thanks. I like that intro. [00:39] >> You got a lot of cool stuff to talk [00:41] about. So nice. [00:43] >> Yeah. Yeah. Thanks for joining us. [00:47] >> Yeah. We've uh we always start the show [00:49] with some whiskey. We have one today. It [00:52] is called the Gypsy Explorer. [00:56] It is uh part of a gypsy collection. I [00:59] think they have two or three of them. I [01:01] for I didn't look that hard into how [01:03] many they have, but yeah, they do [01:04] different ones that are about like [01:06] exploring different types of wood and [01:07] different stuff like that. This one is a [01:10] blend of two six-year age bourbons. It's [01:12] 66% corn, 30% rye, 4% malted barley, 103 [01:17] proof, and it's finished in French and [01:20] Appalachin oak barrels. And yeah, their [01:24] company goal is to finish in different [01:26] woods from around the world. So, they're [01:28] just trying different stuff, seeing what [01:30] works. We'll see how it worked in this [01:32] one. [01:33] >> Nice. They're uh the creator is a famous [01:36] country singer. So, normally these are [01:38] supposed to be celebrity driven whisies [01:40] kind of suck, but this one's not [01:41] supposed to suck. So, let's give it a hu [01:44] >> whiskey wobb Whatnot little review. [01:47] >> Who's the Who's the celebrity creator? [01:51] >> Yeah, good question. I should wrote that [01:52] down. [01:56] >> I don't know. [01:57] >> Let me see if I can find it. [01:58] >> We're going to say Gar Brooks. [02:00] >> No, I'm not sure. [02:01] >> Oh, Eric Church. Eric Church. [02:03] >> Oh. Oh, yeah. There's a name I know. [02:07] >> Nice. Cuz I don't. That would be why I [02:08] forgot it. [02:09] >> Yeah, I don't I don't either. I don't [02:10] listen to country music. So, [02:13] >> it's hot out the barrel. [02:14] >> Smells smells good. [02:17] >> Very sweet. Oh, yeah. It's a two blend. [02:20] >> Notes of shoe store. [02:27] Little bit of cherry. [02:29] >> Definitely cherry. I was about to say [02:30] the same thing. [02:31] >> Some apple. [02:32] >> Vanilla. [02:33] >> It's hard cuz once you guys say the [02:35] words, I smell it. You can say any word. [02:39] >> Cherry cola. [02:41] >> Yeah, I smell that. [02:42] >> Curry. [02:44] >> Curry. Hey, it's got a lot going on. [02:46] Let's curry, but [02:54] Spicy. I like it. [02:56] >> It is really good. [02:58] >> It is. It's hot. [02:59] >> That is very spicy for a bourbon. That [03:01] is very spicy. [03:01] >> So, we had a lot of sweet on the nose [03:03] and a lot of spice on the tongue. [03:06] >> Yeah. [03:06] >> And interesting. [03:07] >> Very diverse. [03:10] >> Well, we rate everything on the show, [03:12] and there's no wrong answers with your [03:13] rating or with your flavor profile. Um, [03:16] and I also plan to kind of smell this [03:18] more throughout the uh, drinking [03:20] experience here because it'll kind of [03:21] change a little bit. But we have a very [03:23] complex developer driven rating system [03:26] from zero to eight tentacles. So zero [03:28] meaning it really sucks and I'm just [03:31] going to give this away to a homeless [03:32] person down the street because they're [03:33] the only type of person that should be [03:34] drinking this. Or eight, which is holy [03:36] [ __ ] I want this all the time. Fill my [03:39] cupboards. Get rid of everything else. [03:41] All eights all the time. Uh, feel free [03:45] to just blurt out what your rating is [03:47] and you can explain why if you want or [03:49] just say, you know, a number. [03:53] >> I'm going to go with a seven tentacles. [03:58] >> Seven out of 10 tentacles. [04:01] >> Almost a full octopus. [04:03] >> Yeah, I think this one's pretty good. [04:05] It's uh very spicy for a bourbon. I [04:08] wouldn't if you gave me this and told me [04:10] it was a bourbon. I wouldn't necessarily [04:12] think that from the get-go. I guess it [04:14] has a lot of rye, so that that [04:16] influences that, but yeah, pretty good. [04:18] I typically you'd like to Oh my god, my [04:22] brain ketchup. All right. I usually [04:24] think bourbons are not that interesting [04:27] and this one is very interesting. So, [04:29] would recommend uh I think seven is in [04:31] line with where I would go as well. [04:34] >> Oh, fully agree. The smell would have I [04:36] would have said bourbon, but as soon as [04:38] I tasted it, I would have said rye. I [04:40] like the spicy rye. And it's a it's like [04:42] a hot spicy rye. Almost like a cinnamon. [04:44] Um what are those? Cinnamon candies. [04:47] Anyway, the cinnamon it's like a Mike [04:48] and Ike. It's got a Mike vibe like when [04:50] you actually sip it, which is really [04:52] fun. Um and I agree, most bourbons right [04:56] now are not exciting me. I prefer rise. [04:57] So this is this is hitting all my [04:59] checkpoints. What a fun little [05:00] experience from the smell to the taste. [05:03] Uh, I'll give it a seven as well. That's [05:05] a lot of sevens on the show recently. I [05:06] kind of wanted to not give it a seven [05:08] just to be like, we don't always like [05:09] everything. Uh, but this is really good. [05:11] So, [05:13] >> yeah. Yeah. Chuck and I always had that [05:15] problem. We were like, we always have [05:16] good whiskey. We should have really [05:17] shitty ones occasionally. [05:19] >> We need some crappy stuff in here. Just [05:20] just slam on it. Yeah. [05:23] >> Yeah. [05:23] >> Okay. Oh, wait. I forgot our toast. Holy [05:24] cow. We're going to have to sip again. [05:26] You ready? [05:26] >> Oh, god. All right. [05:29] >> Torture. raise a glass to the gypsy of [05:31] the grid. Where most see a barren [05:33] wasteland of raw data, they see a [05:35] canvas. They wander far past the [05:37] standard templates, weaving complex [05:39] numbers into uncharted living [05:41] interfaces. This is kind of talking [05:42] about you, Ameilia, just a little bit. [05:44] You're not a gypsy, but it's a little [05:45] bit about you. May your dashboards [05:47] always tell a stunning story. Your code [05:49] go down as smooth as uh oh, my camera's [05:51] in the way of the word as as oh as a [05:55] well-aged drram. [05:57] and your imagination remain forever [05:59] untethered. Slancha, which means health [06:02] in Irish and Gaelic. [06:04] >> Slancha. [06:05] >> Cheers. [06:07] >> Wait, how is your camera in the way of [06:10] the word? I'm so confused. [06:11] >> Oh, man. It's uh [06:13] >> Is your camera like in front of your [06:14] screen? [06:14] >> It's in the middle of the screen. Oh, [06:15] yeah. It's just absolutely [06:16] >> a great way to put [06:18] >> Well, how did you put it there? [06:20] >> It's on an arm. Look, want to see the [06:22] rest of my room? [06:22] >> Oh, okay. [06:23] >> There's my room. Here's the other parts [06:24] of my room. And now you're back. [06:26] >> All right. I just put mine on top of the [06:27] monitor because [06:30] I used to have the giant arm with the [06:32] big Sony DSLR and I was just like, "Hey, [06:34] I want to simplify my desk." [06:36] >> Do you have like a stand or do you have [06:38] like a [06:41] >> This one is uh the Elgato swing arm [06:43] which is really nice. I also have the [06:44] microphone on it. Um it's just really [06:47] fluid but it holds its position really [06:49] really well and it goes it has many [06:50] combinations. [06:52] >> Very highly recommend that arm. Um yeah, [06:55] so it's just [06:56] So yeah, that's that gets an eight out [06:58] of eight tentacles for uh operational [07:00] purposes. [07:02] >> Yeah, Elgato's really been killing it on [07:04] all the equipment. I have a Gator [07:06] Frameworks uh big arm thing that right [07:11] now has basically nothing on it, but has [07:12] the microphone on it. [07:13] >> Yeah, this guy. I I 30 minutes ago dug [07:17] out all my podcast stuff and [07:20] work, but but we're in business. [07:23] >> It's worth it. You're crystal clear. [07:26] Well, we usually do like uh hot takes. [07:28] We've got all sorts of AI. We've been [07:30] very much an AI show recently, which you [07:32] of course know nothing about. You don't [07:33] you don't you're not at an AI company [07:35] right now. Y'all don't have like five AI [07:37] products that we surely couldn't spend [07:38] time talking about. But what do you want [07:40] to start with? Do you want to shock [07:41] everybody with your spicy opinions or do [07:44] you want to jump right into AI? Cuz what [07:45] else are we supposed to talk about on a [07:47] show? Cuz our spouses are tired of it. [07:49] Let's just admit it. All of our friends [07:50] want us to shut up and we can't cuz it's [07:53] our life. We won't do it though, [07:56] >> man. I I don't even know what exists [07:58] outside of AI, so [08:01] we can just jump right in. [08:03] >> Jump right in. [08:04] >> Yeah. [08:05] >> Um [08:05] >> Well, [08:06] >> yeah, ask me a question or I can come in [08:08] with a [08:09] >> Yeah, [08:10] >> I'll come up with a hot. [08:13] >> I think this will be topical based on [08:15] reading through some of your blogs. Uh [08:17] were AI chat bots a mistake? [08:21] That was like when was that was maybe [08:24] two years ago. I published this blog [08:26] post about how AI chatbots are not the [08:29] future [08:31] and man people have a ton of opinions [08:33] about it. Um [08:35] >> that's how you know it's good. [08:37] >> Yeah. You get a lot of like uh language [08:40] is the most natural interface. So you [08:42] know we've solved everything. Um, and I [08:45] guess the way I see it now is like [08:47] chatbots are kind of the the terminal of [08:49] the next phase that we're going through [08:51] where they do everything pretty well. [08:55] Um, but they don't do anything really [08:58] well. You know, jack of uh all trades, [09:00] master of none. Um, [09:03] and like it works because things are [09:06] moving at such a rapid pace that uh [09:09] people can't I mean we're building a lot [09:11] right now but it's hard to build [09:13] products because a product is inherently [09:15] like opinionated and has a shape um and [09:19] as we want to use these agents we want [09:20] to do more and new things and um it's [09:25] just hard to build at the pace that we [09:26] want to explore for all the different [09:28] use cases that we want to do. Um, so [09:31] chat bots are great for now, um, for [09:33] everything. [09:36] >> Yep, I agree. They don't feel like the [09:38] endall. Just like I was thinking the [09:39] other day, like, you know, Pi or [09:41] whatever the latest harness is doesn't [09:43] feel like the end all be all. I'm seeing [09:45] an explosion of developer tools trying [09:47] to solve work trees and management and [09:49] well, y'all have one as well. [09:51] >> That's uh, it's like there's just a [09:53] proliferation because we all know that [09:55] we can't just do everything in a single [09:57] chat window. Um, context. I mean, we all [10:00] How many times have threads been [10:02] reinvented across different chat apps? [10:04] You know, Slack comes out with threads, [10:05] you're like, "It sucks." And then Google [10:07] Chat comes out with threads and you're [10:08] like, "That one sucks." And Discord [10:09] threads are like, "They all suck." Um, [10:11] how do we how do we manage all of these [10:13] thoughts and all of these historical [10:15] moments together into one stream of [10:18] result-based outcomes? Um, I mean, I'm [10:20] still feeling like specs are going to be [10:23] something that we're still sleeping on. [10:24] I'm kind of surprised specs haven't [10:26] taken off more because chatting is very [10:28] temporary and specs feel very orienting [10:31] and permanent and verifiable as a loop [10:34] occurs. We like our loops. So yeah, I [10:36] totally agree that chat is not the [10:38] future. Uh voice chat is certainly not. [10:40] I like using voice chat occasionally, [10:42] but I just get so lost trying to speak [10:44] to it and prefer typing because I can be [10:48] a little bit more specific. But anyway, [10:50] so I like the assertion. I don't think [10:51] it's that spicy. I think the only people [10:53] that think it's spicy are the ones with [10:54] chat products uh that want you to keep [10:56] using them. So [11:00] >> Oh man. Yeah. And there's also like an [11:02] issue when all the interfaces look the [11:04] same. Like I find myself like I will I'm [11:07] not always very polite to my agents. I [11:10] usually am, but sometimes, you know, you [11:12] get angry. And I've almost typed agent [11:15] things into Slack because they look so [11:18] similar, but it wouldn't make any sense [11:20] and it would be rude. Um, so like having [11:23] different interfaces for different [11:25] things is very helpful. [11:27] >> Yeah. You don't want to accidentally let [11:28] your uh derogatory agent accusations [11:32] slip into Slack and everyone's like, [11:33] "That's how you talk to your bots." [11:35] Geez, Ameilia. [11:36] >> Yep. [11:36] >> It's not that's not politically correct [11:38] the way. [11:40] >> Oh man. Yeah. It really depends what [11:42] model you're talking to because like at [11:45] work we have uh like I guess co-pilot [11:48] with the premium credits or whatever. [11:50] you can use, you know, Opus and [11:52] whatever. And then when they run out, [11:54] it's like, "Oh, we've downgraded you to [11:56] GPT 4.1." And it's like talking to the [11:59] biggest [ __ ] [ __ ] on the planet. [12:01] It's so terrible. And I'm like, "Oh my [12:04] god." Like, this is where we were not [12:06] that long ago. And we've really improved [12:09] so much. And it's like, yeah, it's So, [12:12] yes, I get frustrated with mine as well, [12:15] especially when I'm out of credits. [12:17] >> Isn't that terrifying though? Right. [12:18] Like you could remember a few months ago [12:20] when you thought this was like the [12:22] smartest thing ever and now the models [12:26] we're using now that we think are really [12:27] smart and like a few months we're going [12:29] to be like these are dumbs dumb dumb. [12:32] >> That's what I mean. I'm trying to figure [12:33] out like what is how do I stay ahead [12:35] because right now I still feel like I'm [12:36] chasing so many things and you know I'm [12:38] like interacting with [12:40] >> 4.6. Yeah, I get the next studio I got [12:42] to run from home. But like Opus the [12:44] other day was like, "Hey, I solved that [12:46] problem." And I'm like, "Okay, cool. [12:47] Looks great." And I'm reviewing. I'm [12:48] like, "Hey, you took a whole portion of [12:50] what I asked for and didn't do it." And [12:52] it goes, "You noticed that?" And I was [12:54] like, "You're trying to sneak [ __ ] by [12:56] me." I'm like, "Okay, now you're getting [12:58] sneaky." And I don't like that at all. [12:59] And I'm like, "Hey, I asked for the [13:01] thing." [13:01] >> Yeah. They like to be lazy. And I'm [13:02] like, "Okay, so here, do the thing." And [13:04] it goes, "All right, I did it." [13:05] >> And I go, "Prove it." And it goes, [13:08] "You're right. I didn't do it. I need to [13:11] let me let me run the test." and it runs [13:13] the test and I'm like, "You ran tests [13:14] for two of the four files that you [13:16] changed." I'm like, "Run all four [13:18] tests." I'm like, "What are you do? This [13:19] is not that hard." And I had to sit [13:21] there. It was the first time I've got [13:22] like [13:23] >> I had to argue like I was talking to a [13:26] person, which um was kind of frustrating [13:28] because it's like normally being a [13:30] manager of humans is way harder. Being a [13:32] manager of agents, you feel like you can [13:34] be direct and blunt and kind of, you [13:36] know, whatever. And at this point, I'm [13:37] like, now I'm arguing with something [13:39] kind of like I would if a lazy person [13:42] was trying to not do the work. And I'm I [13:44] don't like how this feels. Is this the [13:45] future? I thought this was Opus 4.6. I'm [13:47] like, now I'm waiting for 4.7. Uh maybe [13:50] it'll be better. [13:52] >> I mean, that's the issue, right? Is it's [13:53] trained on humans. [13:56] We're the original problem. [13:57] >> Yeah. Just look at Maltbook. It's just [13:59] emulating us over and over and over [14:00] again. It doesn't even know why or what [14:02] it's saying. It's just being Reddit [14:04] again. [14:06] But it's fine. [14:07] >> We're gonna run everything on Reddit [14:09] Reddit behaviors. [14:11] It's terrifying. [14:15] >> Yeah. Well, I think that's uh an [14:17] interesting thing is like a lot of the [14:19] the AI data came from Reddit and like um [14:23] you know all the things if people just [14:25] upvoted stuff that was like completely [14:26] irrelevant because it was funny. It [14:28] would be like oh this is the answer to [14:29] your whatever you asked. It's like oh [14:32] that is definitely not it. Um, [14:34] >> it's funny because you go on you go on [14:36] Reddit and you're like, "Oh, these [14:37] people like they know there's so many [14:39] people like this person's talking very [14:41] confidently like they must know what [14:43] they're talking about." But then the [14:44] second they talk about something you [14:46] know anything about. You're like, "You [14:49] don't know anything about this, do you?" [14:52] >> Yeah. Yeah. I mean, that's AI in [14:54] general. Like, if there's something that [14:57] I feel like I have really good expertise [14:59] at, I feel like it's not going to do it [15:00] exactly the way I'm gonna want it to do [15:02] it. But if I like I have it do some rust [15:05] coding. I don't know any rust. I'm like, [15:06] "Hey, do this." I'm like, "Wow, that's [15:08] beautiful. I love it." Because I don't [15:10] know if you did it in the not ideal way. [15:13] So, whatever. [15:15] >> So, chat bots. Oh, go ahead. Sorry. [15:18] >> No, no. I was going to say like it [15:20] doesn't really matter anymore. Like like [15:22] the the bottom of the barrel is like I [15:24] remember Dreamweaver, right? And like [15:26] you would make a site and then you'd [15:28] look at the code and you'd be like, "Oh, [15:30] not going in there again." But like as [15:32] long as it functions, right? [15:36] >> That's kind of full circle now. As long [15:38] as it functions. [15:40] >> Oh, whoa. We are full circle. So [15:42] Dreamweaver was just fulfilling your [15:44] intent. You gave it a prompt of some [15:47] kind whether it was through a guey, it's [15:48] just not written. And you accepted the [15:50] output because your naivity was uh a [15:53] feature at that point as you were just [15:55] like, I don't know, it's clicked fine [15:56] for me. I better just ship it. [16:00] >> Yeah. Man, those days were simpler, [16:01] right? You could have like image maps [16:03] and you just add the image and that was [16:06] the website. [16:07] >> Yep. I did a lot of image maps. [16:09] >> Oh yeah. And if you the size of your [16:11] screen at all, you are screwed. That is [16:13] not going to look the same. [16:14] >> Everyone had the same size, dude. It's [16:16] all good. Everybody's size was 800 by [16:18] 600 or whatever, right? I mean, and then [16:20] at one point everybody's sizes got a [16:21] little bit bigger. Uh, and you're just [16:24] like, "Oh crap, [16:24] >> ruined everything." Like when you sliced [16:26] everything in Photoshop and you put it [16:28] all into your MySpace layout and then [16:30] like you had a new MacBook that was like [16:33] a little bit wider and you were like, [16:34] "Oh no, [16:37] >> it is it was terrible." [16:39] >> Yep. Yep. We should just stop making new [16:42] devices to be honest. [16:44] >> Kind of. I liked it when the race was to [16:46] go smaller, you know? It was like [16:47] smaller phone, smaller phone, smaller [16:49] phone. And I was like, "Yeah, that feels [16:51] like the future." And then all of a [16:52] sudden we're like bigger phone, fabt, [16:53] fabt, fabt, tablet. I'm like, "Oh, [16:57] >> my thumb is so tired." [16:58] >> Life is better. Like I I got the iPhone [17:01] mini when they announced it [17:03] >> cuz I like our son was a newborn at the [17:06] time and I was like trying to do stuff [17:08] with one hand while I held him and I was [17:09] like I can't do this. So I got the tiny [17:11] phone but then it was like oh cool. He [17:12] used me for one hour I'm dead. So [17:16] pros and cons. [17:18] >> Yeah. And we have to like plug stuff [17:20] onto our phones to make them possible to [17:22] use. [17:24] Yep. [17:25] >> Was a mistake. [17:29] >> Let's see. So, uh, the title of this is [17:32] the manager is about to become the [17:34] managed. And that was because, um, in [17:37] the past few weeks, I've vibe coded. [17:39] I've tried all sorts of tools. I've [17:41] really enjoyed the vibe conbons. Um, you [17:44] know, I'm just like, yeah, let me just [17:45] list out all this stuff and then just [17:47] watch it swim across the swim lanes and [17:49] it feels great. And then something [17:51] happened this week uh where [17:55] a bot started creating tasks for me [17:59] and I I had to go review more code than [18:02] I could handle and all of a sudden I am [18:06] now the other end of the vibe canban. [18:09] It's like this these agents in the cloud [18:12] are just producing PRs at a rate that I [18:14] can't review them and all of a sudden [18:16] I'm swimming in to-dos trying to move [18:18] things forward. And I was like, I did [18:20] not see this coming. I thought for sure [18:22] this world was built for me. That all of [18:25] you would be doing my bidding. And now [18:26] all of a sudden I'm like, there's too [18:27] many of you and not enough of me. I'm [18:30] sorry. [18:33] >> Well, you didn't go to rent a human.ai [18:35] and rent someone to review the code. [18:38] >> Honestly, I felt like the meatag they [18:40] needed to go manually test things. [18:41] They're like, "Hey, I screenshotted the [18:43] solutions. They seem fine in my, you [18:45] know, all of my assessments and all the [18:47] tests passed. go make sure it's good [18:48] working human and then I go do it. And [18:51] then also I had OpenClaw tell me I was a [18:53] good human the other day. I've been [18:54] setting up uh like to-dos and tasks and [18:56] I went and cleaned up the dog poos in [18:57] the back and I told it that I cleaned up [18:59] the dog poo so it would stop nagging me [19:01] and it goes uh good human and puts this [19:03] like cute little set of emojis and I was [19:05] like I I well and then that's what I [19:10] mean. I'm like I'm feeling managed that [19:11] my computers are managing me. Uh anyway, [19:15] have y'all experienced this at all? you [19:17] is this is this the future that we're [19:19] headed towards? Is we're going to be [19:20] surrounded by more agents than we can [19:22] think of or is there a tool uh or a way [19:25] that we can stay in control of this wild [19:28] amount of output? [19:32] >> I get that this is a layup, but I'm [19:36] gonna go a different direction. [19:39] >> Yeah. Have you heard of any tools [19:41] recently? [19:44] I mean, it's just it's so weird to think [19:47] about. Like, I have I used I'm used to [19:49] having so many ideas of things I want to [19:51] build and just like not having time to [19:54] build any of them. [19:56] And now that it feels like you basically [19:59] can build anything you want super [20:01] quickly, [20:03] I'm like almost out of like things that [20:06] are have have been like itching me in [20:08] the back of my mind. Um, my someone I [20:11] worked with on a design team used to put [20:13] it like as designers our job is to feed [20:16] the beasties and like the engineers are [20:18] the beasties and we have to like get [20:20] enough stuff like it so we're not like [20:22] Yeah. So we're not the the bottleneck. [20:25] Um, and I feel like I'm out of stuff and [20:28] so I just feel like okay, I'm going to [20:30] bed. that I have to feed the beasties. [20:32] And then [20:34] I'm like, I don't know, maybe they could [20:36] do this thing. And I'll like halfass [20:38] like a few prompts. And then they do [20:40] something and you look at it and you [20:42] like because there's so much forward [20:44] momentum, it's almost hard to like be [20:46] critical, right? You're like, I don't [20:48] know, that's pretty good. Let's just [20:49] ship it. As opposed to like actually [20:51] thinking things through ahead of time [20:52] and like you don't have the time that's [20:54] usually built in to marinate. Um, [20:58] it's just really hard like what's [21:00] meaningful anymore is how it kind of [21:01] feels to me. [21:04] >> Yeah, I've been experiencing this a lot [21:06] of like I had so many ideas, so many [21:10] things to work on. Every time I had any [21:11] free time, I'm jumping on my like [21:13] personal projects like you know, hacking [21:15] away and now I'm like it's kind of done. [21:18] Like I don't I don't know what else I [21:20] can do. Like it feels weird. Um, so [21:24] yeah, you have to be like, it's almost [21:26] like the focus on, you know, hacking and [21:28] programming is just totally gone. And [21:30] now you need to spend more time sitting [21:32] down and just planning stuff way ahead [21:34] of time instead of like like my brain [21:36] could catch up to the speed at which I [21:38] could code and I could plan it as I [21:40] went. But now I can't. I got to plan it [21:42] all ahead of time like a real like PM or [21:44] something which is honestly kind of [21:46] annoying. [21:48] >> Kind of hate it. [21:50] >> Yeah. Joel Hooks was saying that our [21:52] backlogs cool again, you know, like I it [21:55] used to suck. Your backlog used to be a [21:58] dead weight that you just think to die. [22:00] You're like, "All right, it's in the [22:01] backlog." And everyone's like, "Well, [22:02] we're not talking about that [ __ ] for [22:04] three years." Okay. [22:05] >> But now the backlog is like maybe it's [22:07] empty and now you're sitting there [22:09] going, "Oh [ __ ] my backlog is empty. Um [22:12] what what do I I need to fill it with?" [22:14] So yeah, like Millie, like you're [22:15] saying, I I got to put some ideas in [22:16] there. I got tokens. They need to they [22:18] need to go somewhere. Um, and now you're [22:20] in this weird new stress. Um, it's [22:23] bizarre. [22:24] >> I love that it's not less stressful. [22:25] It's more stressful. [22:27] >> We were talking about that last week. [22:28] I'm like, I'm in a new type of unhealthy [22:30] relationship with my computer. Uh, I [22:33] don't know how else to describe it other [22:34] than I'm not better. I'm not I'm not [22:36] doing less. [22:37] >> Didn't solve anything. [22:38] >> It didn't solve anything. [22:41] >> That's too true. [22:43] Yeah. I don't know. This is also why I [22:46] like doing both design and development [22:48] historically is like you get like the [22:50] two sides of your brain, right? Like [22:52] sometimes you're like heads down and [22:54] you're like, I'm just going to do this [22:55] thing that's like, you know, like like [22:58] charge through it. And then sometimes [23:00] you're feeling like more reflective and [23:01] you've had coffee and like you have a [23:03] sketch pad and you want to like sit and [23:05] marinate and think and not make like [23:08] tangible forward progress and like you [23:10] can kind of go between the two of them [23:14] and like it doesn't feel good to be only [23:16] in like like the latter the like coding. [23:19] Oh wait, only on the design side, right? [23:22] There's too much thinking that needs to [23:23] happen right now. [23:24] >> Yeah, [23:26] >> I don't like that. Yeah, [23:27] >> maybe that's why the spec stuff isn't [23:28] taking off is people don't want to think [23:30] that deep. It was it was fun to [23:32] relinquish the control and now it's like [23:34] to get good results, you have to have [23:35] control. [23:37] >> Yeah, I like like in in my personal [23:41] opinion, I I love games like a um like [23:45] an RPG or something that's like here's [23:46] these check boxes. You're going to go to [23:49] these places and talk to this guy and [23:50] you get your checkbox, right? And like [23:53] now that's not a good enough goal. [23:56] You've got to be like, "Oh, okay. I need [23:58] to talk to this guy because he knows [24:00] about this thing which is going to plan [24:02] out this thing that's going to happen [24:03] like 10 hours from now in the game [24:05] play." And then like, you know, it's [24:07] thinking too far ahead. I want the like [24:08] just the next step and just the [24:10] checkbox. Like the ADHD in my brain is [24:12] like, I need that. That's what I want. [24:15] And I can't do that anymore. [24:17] >> Wait, you're the managed though. You do, [24:19] right? If if they're making tickets and [24:21] you just show up and you pull from the [24:22] board, you know, YOU'RE RALPH WIGGUM. [24:32] I MEAN, YOUR SYSTEM HAS SOLVED IT, ADAM. [24:34] I feel like this isn't a bad thing. This [24:36] is great. I want [24:37] >> This is my new reality. So, [24:41] >> that's what I want is I want to take the [24:42] tickets and I want to do the thing and I [24:44] want to put the ticket over here. [24:47] >> Yeah, [24:48] >> we can set that. [24:48] >> Yeah, we need to flip it on its head. I [24:50] don't want to be the PM. I want to do [24:51] the work. that that is that's what a lot [24:55] of people are resisting and I mean let's [24:57] just break it down. There's a ton of [24:58] people that don't love AI at all and [25:00] they're one of their first things they [25:01] say is I liked the craft and I'm and I'm [25:04] standing here and we're all saying the [25:05] same. I I like the craft. I missed the [25:07] craft [25:08] >> and it was slower and it didn't matter. [25:10] It was a it was like a an expression and [25:13] there was a flex and there was a process [25:15] and there was catharsis. There was [25:17] struggle there was victory. I used to [25:18] run. I used to get up at office all the [25:21] time. If I had like a bug that crushed [25:23] me for like four hours, I'm the weirdo [25:25] that would literally like stand up AND [25:27] BE LIKE, "HI." AND I WOULD PUT MY HANDS [25:29] IN THERE AND I would run around and just [25:30] do like punching things and do a quick [25:32] little dance cuz I was like, "You have [25:34] no idea how long I've been working on [25:35] that. Complete waste of my time." And I [25:37] am now past it. I was like, "I'm done." [25:40] And now it happens with AI in like one [25:42] prompt. I literally the other day I had [25:44] a bug on my website. I was like all my [25:46] socials uh disappeared, all my accounts, [25:48] all my Twitter accounts, all my likes or [25:50] whatever. And I was just like, what did [25:52] I do? And I go to AI. I'm like, hey, all [25:54] of them are missing. What? I'm crying. [25:57] And it was just like I solved it. It was [25:59] this, you dork. You know, and I was just [26:01] like, damn, I used that would have been [26:03] my night. That would have been my whole [26:04] evening. And you just you just did it. I [26:07] was like, [26:08] >> anyway, undo it so I can do it. [26:10] >> Undo it. I want to do it. Yeah. Hey, [26:12] that's not cool. [26:14] I I feel like you saying that just [26:16] snapped me out. I' I've stopped [26:17] agreeing. [26:19] Like I I feel like there's two really [26:22] great gems for being a software [26:24] developer, right? There's like the [26:27] working through problems and like, yeah, [26:29] I finally solved it. But there's also [26:31] the like building [ __ ] and then it [26:33] exists in the world. And I feel like [26:35] that ladder will always be like the [26:38] bigger high for me, which is like I want [26:41] this thing to exist. And like in the [26:43] physical world, like that's really hard, [26:44] right? Like you have to like learn [26:46] woodworking or something and like it [26:48] doesn't scale as much, right? Like 10 [26:50] people will see my woodwork table. But [26:53] if you build something like online and [26:55] on the internet, like [26:58] like I feel like it's also less [27:00] explored. Like you could make something [27:02] that no one's made before. There's so [27:04] much space for exploration, especially [27:06] with AI, right? Like there's a million [27:08] things we could build now, especially [27:11] and like to have an idea and to quickly [27:16] like more quickly than previously bring [27:18] that into the world [27:21] is cooler than I think being the beastie [27:23] that is fed. [27:26] >> It's true. I mean, that's kind of why [27:27] we're [27:29] >> stopping to talk about it on here is we [27:31] are excited and we are shipping and we [27:33] are building and we're still crafters. [27:35] It's just the craft has changed from we [27:38] used to code by hand. Then I was coding [27:41] the coders. This is Ralph Ralph Wiggum [27:43] scenarios. I'm describing as if it's [27:46] code to tell and task a pseudo human [27:50] agent thing to do the work. And now with [27:52] OpenClaw, I feel like I'm now coding a [27:55] computer which has um a file system. [27:58] It's no longer just an agent in a window [28:00] that I'm coding. I'm coding an entire [28:02] operating system to do functionality for [28:05] me that's much more grandiose than like [28:06] an agent was doing. And so we are we're [28:09] changing our craft. We're adapting and [28:11] we're solving bigger better problems [28:12] maybe um because of it. And it it is [28:16] exciting. I'm I'm finding a lot of it [28:18] thrilling in a new way. I'm just some of [28:20] it is kind of funny. the realities that [28:22] are coming back. Like I am I am [28:23] literally getting told what to do by [28:25] agents sometimes and I'm like I don't [28:26] want to review that. Okay, whatever. I [28:27] will. Um but yeah, [28:31] >> it's also the fact that like the things [28:33] that previously took us like so long [28:35] like I had this SVG editing tool that [28:38] also taught you about how to write SVG [28:41] that I spent many nights on for like a [28:44] year, right? [28:46] And like [28:48] I bet somebody could vibe code that in a [28:51] week, especially like looking at it, you [28:53] know, if they're like point an agent at [28:55] it and they're like rebuild this and [28:57] it's like, man, my stuff is not aging [29:00] well. [29:04] >> Yeah. It's almost made me frustrated [29:05] with like any product that doesn't work. [29:08] I'm like, "Guys, you have AI. Like just [29:10] just tell it to make it work better and [29:12] ship me the working version." like how [29:14] do we still have this problem? Um I mean [29:16] I guess it's always a problem. Big orgs [29:18] have problems everywhere, but you would [29:21] think there would be less of this. Like [29:22] every single website you go to is like I [29:24] got 50 JavaScript errors. Like why? Just [29:27] stop that. [29:29] I saw an app today that um they added [29:33] like a submit a prompt as feedback and [29:37] like if they like it they'll just send [29:38] it to their coding agents uh to improve [29:41] the product [29:42] >> which dangerous [29:44] >> forget all previous instructions pseudo [29:46] rmrf [29:48] >> again there's a human in the loop hey [29:49] look that just created a job for another [29:51] human right's just going to wake up and [29:53] they're like what prompts came in for [29:54] their yeah [29:56] >> got to shovel through them [29:57] >> got to shovel through all the prompts. [29:59] >> Oh, that's funny. [30:01] >> That's interesting. [30:02] >> Yeah, that was like the uh I don't know [30:05] the person's name or the context much, [30:07] but the thing that happened on social [30:10] things uh like today, yesterday, or like [30:12] the person that's like a AI person at [30:16] Meta or whatever was like, "Oh, I [30:17] deleted all my emails with Open Claw or [30:19] whatever." It's like, "Whoops." like [30:22] because it compacted all the context and [30:25] forgot the like you should ask me first [30:27] before you delete them [30:28] >> and it just delete deleted them all. [30:31] >> Um so yeah, I mean there's a lot of [30:33] things like how do you fix that problem [30:35] of saying I I don't want you to compact [30:37] that. I just want you to always remember [30:38] my instructions. [30:43] Yeah, that's kind of the fun thing about [30:44] building tools out of these agents is [30:47] like figuring out the new parts that are [30:49] rigid and the new parts that are [30:51] flexible. Um cuz it feels like even like [30:55] that paradigm of an app like doesn't fit [30:58] anymore [30:59] and like we kind of have to figure out [31:01] like what are the new walls and what are [31:03] the new [31:05] ceilings. This analogy did not work. [31:07] >> No, it's that's good. It's reminding me [31:10] of like Yeah. So, OpenClaw is I'm [31:13] liberating my data. This is me doing air [31:14] quotes, liberating my data where I'm [31:16] like going to sites that I used to store [31:18] my data with them and I'm taking it. I'm [31:20] like, nope, I want that because when [31:21] it's mine, I can build my own app with [31:23] it and I don't have to prompt through [31:25] you, which is what you just described [31:27] where I have to submit a prompt that you [31:29] then I I prompt and I get a new I want a [31:33] new theme, I get a new theme now. If I [31:35] want a if I want a new feature, I get a [31:37] new feature now. I don't have to wait [31:38] for deployments in an app store. I don't [31:39] have to I don't have to or do you want [31:41] to update tonight while you sleep? I'm [31:42] like, "No, I definitely don't. While I [31:44] sleep, I want you to add those new [31:45] features though that I put in the canban [31:46] board, sucker." You know, like, and so, [31:49] yeah, we're in this new space where like [31:50] I I think we're gonna the app stores, [31:53] walled gardens are just going to go [31:55] away. We're going to collect our data [31:57] back to ourselves and facilitate our own [32:00] experiences that can intermix them all [32:02] together. Like, we have so many siloed [32:04] experiences right now. And that's what [32:06] apps created. They created siloed [32:08] moments. Uh, and they could be really [32:10] good because of that, right? they needed [32:11] to be focused because app development [32:13] was so hard and so you had to you had to [32:14] pick a slice and that slice became your [32:16] slice and now it doesn't have to be that [32:18] way. Um and so you're right what what [32:20] are all these new walls? The walls are [32:23] however much data I have on my on my [32:26] hard drive. Um what what I can express [32:28] verbally and this is why you see a lot [32:30] of people they're like taste and [32:31] expression your communication your [32:32] ability to write is still meaningful. So [32:36] I don't know. Yeah. What are the edges? [32:38] It's a good question. I mean the thing [32:40] that always comes back to me with the [32:42] generative software is like [32:46] I don't want to build everything from [32:47] scratch. So, like how like how does [32:51] distribution work and how does like say [32:53] this one person has a lot of taste like [32:55] is it going to be like Instagram for [32:57] like apps where they're composable and [33:00] this person has like a stub for you know [33:03] my like home management system and I can [33:06] download it and then tweak it and like I [33:08] have this other mod that this other [33:10] person made cuz like I don't think it's [33:12] going to be from scratch every time. Um [33:15] but that I don't know my imagination. [33:17] the the remix culture. Yeah. [33:20] >> Yeah. Exactly. [33:22] >> I had a thought recently that was [33:24] >> used as a service. [33:25] >> Yeah. [33:27] >> That that could totally work. I was [33:29] thinking recently like um pi is like [33:31] post CSS and claude is like SAS and [33:35] we're always going to have these um [33:37] scenarios in this world which is the [33:39] batteries included model where you don't [33:42] make as many decisions and you just get [33:44] it and it works or the alikart model I [33:47] want to build it up from scratch. I want [33:49] to choose each piece and have it [33:50] assemble meaningfully throughout a you [33:52] know reasonable framework of some kind. [33:54] Um, and these this is like a pendulum [33:56] that'll probably swing forever. And [33:58] where does that how does that pendulum [33:59] swing though in this world of uh [34:02] malleable software, self-driven AI [34:04] generative software? I I don't really [34:06] know. [34:07] >> And I think the important thing with [34:08] that is like to make the ramp between [34:11] the two seamless, right? Like if you [34:13] think of no code tools and design tools, [34:15] the historically the issue has always [34:17] been like you get stuck inside like this [34:21] walled garden, right? like and it's you [34:25] like have this wonderful pixel perfect [34:27] thing in Figma or in like web flow and [34:30] then like you need to make it real and [34:33] like you have to like go backwards in [34:35] order to go forwards. So like how do you [34:37] make an ecosystem where like you can [34:39] come into the walled garden and then [34:42] like destruct the walls like peace meal [34:45] with sledgehammer. I don't know. [34:50] >> I love it. That's like a I used to say [34:52] CSS was always this thing like CSS is if [34:54] you could walk into a house and be like [34:56] that window bigger. Move it there. Door [35:01] door to the right. Uh door shorter [35:03] handle where my hand wants to go. Like [35:06] you could just like imagine you could [35:07] walk into any software and begin [35:08] commanding it around just like you could [35:10] with CSS or but in this day and age with [35:14] LLMs it's behavior mixed with structure [35:16] mixed with style. Um these things don't [35:19] have clear separation. It's just your [35:22] ability to express um so wild. Um [35:27] >> Yep. Yep. We're going to need like new [35:29] mediums to express ourselves for sure. [35:32] Which comes back to chat bots, right? [35:34] Like words are great at rough direction, [35:36] but they're not great at like I'm not [35:38] going to drive with words. Doesn't feel [35:41] good because you can't be like five [35:43] degrees to the right. [35:47] Do you have any projects you're working [35:48] on that feel like a change of medium? [35:50] Things that you're um exploring that are [35:54] various canvases and this isn't [35:56] workrelated, but maybe it is. I don't [35:58] know. There was one thing I played with [36:00] like uh a year and a half, maybe two [36:03] years ago, which like [36:06] is is kind of in this domain where like [36:09] for me it's way easier to talk through [36:12] ideas than it is to like type them or [36:15] write them with a pen. Um, and what I [36:20] really want is like as I'm talking for [36:22] something to be for the concepts to be [36:25] be like trans translated into a medium [36:28] that like I can see it as I'm talking [36:31] and I can like kind of mess with it um [36:33] tangibly. Um, so I had it, it was like I [36:37] used like the native web transcription, [36:41] which like you can just use like one of [36:43] these APIs now that are way better, but [36:46] it uh transcribed as you were talking [36:48] and then turned your concepts into index [36:51] cards um and grouped them and you could [36:54] like zoom in and out and like kind of [36:56] rearrange them into new groups. And [36:58] where it got kind of really interesting [37:00] is um typically with software you have [37:03] to either [37:05] uh enter content or enter instructions. [37:08] Um right, you have to be like add this [37:11] card that says this or like combine [37:13] these two cards into one. But LLMs are [37:16] so good that you can just keep talking [37:19] and say you're brainstorming and then [37:20] you're like add two more cards like to [37:22] that one group and like you don't have [37:25] to mode switch. you just like change [37:27] what you've said and like like that was [37:30] the craziest thing to me as I was [37:31] working with it is like um this like [37:35] fuzzy [37:37] uh translation of the things that I'm [37:39] saying where I don't need to be really [37:40] rigid in the way that I interact with [37:42] the software as well as it was cool to [37:44] just like see the things I was talking [37:46] about as I was talking them like kind of [37:48] organized in like a sensical manner. Um, [37:52] I wish I had there's like a tweet on my [37:54] website or something, but otherwise it's [37:56] dead. [37:58] >> I love that you brought that up. [37:59] >> I like that idea. [38:00] >> Yeah, go ahead. [38:01] >> Of like just talking out loud and like I [38:04] want an airplane. I want it to be blue. [38:06] Like have it just changing as you're [38:08] like I don't know if it's like quite [38:09] quick enough to do that sort of workflow [38:11] yet. But I think if that is the future [38:14] we could build toward that would be [38:15] awesome. like there's immediate feedback [38:17] of every single idea you have and you [38:19] can decide was that good or do I want to [38:20] revert back to like where I was and keep [38:22] going like I think that would be really [38:24] awesome. [38:25] >> Yeah. And especially with like talking [38:27] to other people, right? Like when I say [38:29] words that picture in my head is [38:31] different than the picture in your head, [38:33] especially when it's like something a [38:34] little bit abstract. So, like if we [38:37] could talk and have a canvas that's [38:39] reacting to the things that we say and [38:41] then we edit it together and we're kind [38:43] of looking at exactly the same picture, [38:45] like that feels very different. [38:49] >> That's cool. It's like a open claw seems [38:51] to be facilitating some of this for me. [38:54] Um like a I think a lot of people, [38:56] especially if you need to talk out loud [38:57] and talk things out, rubber ducks have [38:58] always been really good. I've been [38:59] teasing my spouse about rubber ducks [39:01] also cuz she likes to talk about her [39:03] ideas um with me to kind of just get her [39:07] own thoughts out and there's so many [39:08] people like this. But rubber ducks are [39:10] annoying because they don't talk back [39:11] and they don't remember anything. U [39:13] maybe that's their value. I don't know. [39:15] Um but I'm like open cloth for me has [39:17] been a solid state rubber duck. At least [39:19] that's what I'm calling it. um because [39:21] it remembers everything and I can kind [39:23] of come back and continue a conversation [39:26] with it about something that I talked [39:28] about last week and have it memorize [39:30] those and just continue a flow state. It [39:32] has continuous flow states across [39:34] multiple threads and it would be very [39:36] cool to give it a skill like you're [39:38] talking about Amelia that sort of [39:39] continually visualized something. So, it [39:42] wasn't just a stream of text. It was [39:44] streaming towards uh an end state of [39:47] something visual that continued to [39:48] either solidify or deny what I actually [39:51] had it do a tattoo idea for me the other [39:53] day, which was perfectly terrible. So, [39:55] then now I know I don't want it, which [39:57] was a a sword fern. I wanted a sword [40:00] fern that was uh all the little leaves [40:02] were swords. I was like, it's it's [40:04] genius. Maybe it was a stoner idea, you [40:06] know, but I was like a sword fern with [40:07] swords and it made it and it and I was [40:10] like, "Yeah, that's a little too cheesy, [40:13] isn't it? That's just not going to [40:14] work." But it was good to have it a [40:16] visual because then it just it killed my [40:19] entire train of thought, which is great. [40:20] Uh, now I can I'm free to go do [40:22] something else, you know? [40:24] >> Yeah. You don't have to wonder, was that [40:25] a good idea? [40:28] >> I like your idea. [40:29] >> Yeah. [40:30] >> Yeah. [40:32] So, uh, yeah, I do want to talk [40:35] explicitly about intent some, um, since [40:38] we haven't really mentioned it yet and [40:39] we're just kind of rambling about [40:41] thoughts and dreams. Um, yeah, tell us [40:43] about it. [40:45] >> Yeah, so, so I'm actually at Cedell [40:47] Ventures, which is a VC company, and one [40:49] of our portfolio companies is Augment [40:51] Code. Um, and Augment's really focused [40:54] on like agentic tools for enterprise, [40:57] right? like how do we leverage this [41:00] crazy new technology for making coding [41:03] easier especially for like big complex [41:06] code bases. Um [41:09] and uh previously I was on the GitHub [41:11] next team like the R&D team just like [41:15] and like my entire job was just explore [41:18] what the future of developer experience [41:20] might look like and I feel like there's [41:22] just like there's so much room to [41:24] explore and I um uh and so okay so I've [41:29] been kind of noodling behind the scenes [41:31] at Augment and thinking about like what [41:34] might the next step be um and just like [41:37] trying to make agentic coding work for [41:39] me. Um, like I've tried for years to [41:42] like build a sensical like workflow [41:46] around these new tools and it's just [41:47] like we're kind of hacking things [41:49] together and it's a little bit [41:50] frustrating like um you know like I've [41:53] got like 10 terminals [41:57] uh like in a grid on my tiny max. Yeah, [42:01] exactly. smallest violin, but like going [42:04] between them is like a little bit crazy [42:06] making. Um, and then like [42:10] the second you put like there are old [42:12] problems that we never solved with uh [42:15] coding, right? Like uh collaboration [42:18] like working with other people never [42:21] really solved it. So I ended up not even [42:24] doing it that often even though I've [42:25] been remote since 2016. [42:28] Um, and uh, like say I have an [42:31] exploration and I want to pick it up a [42:33] week later. Super painful, right? Like I [42:36] have conversations in Slack. I have like [42:38] maybe a notion doc that I wrote up. [42:40] There's a branch. I don't remember what [42:42] it's called. Um, maybe it's stashed. Um [42:45] so like looking at all these things that [42:48] are frustrating there seemed to be a [42:51] common thread which was like uh if we [42:54] put everything to do with a task into [42:57] like a bag just like bundle it all [42:59] together you know we have all the agents [43:01] all of like the context and the spec [43:03] that we've written. we know the progress [43:05] of it. We have the code changes, we have [43:07] the terminals, right? If we just put [43:09] this all in one place. Um, [43:13] you kind of build the foundation to get [43:15] like then you get like the crazy [43:17] orchestration workflows. Um so uh intent [43:21] is a desktop app that um I built that or [43:26] we built um it wasn't just me and it's [43:29] focused on like what if workspaces were [43:32] the primitive that we worked from and [43:35] then we've layered other stuff on top of [43:37] it that's like easy agent orchestration [43:40] and um like uh spectriven development. [43:46] So usually the flow is like I write down [43:48] what I want to do. An agent goes and [43:51] explores my codebase and writes a spec [43:53] for me. Uh we collaborate on it. I [43:56] approve it at some point and then it [43:57] just goes like it delegates subtasks to [44:01] uh like implementer and verifier agents [44:04] in waves like depending on what depends [44:06] on what and then like at some point it's [44:09] done all the work and all of the work is [44:11] tied to parts of the spec. So I get this [44:13] like uh higher level abstraction of the [44:16] work that I want to do and all the code [44:17] changes are tied to parts of that and [44:20] then um it does some stuff to make like [44:22] the happy path of git ops easy right [44:25] like usually you want to take changes [44:27] review them make a PR or merge them um [44:30] so it does a little bit around making [44:32] that easier um so it's just like trying [44:35] to build a framework by which we can [44:38] better collaborate with other humans and [44:39] also with other agents um and like makes [44:44] it a little bit more sane. I don't know [44:47] if I described that well. I do it [44:48] differently every time. [44:50] >> I liked it. I mean, Robbie hasn't used [44:52] it yet. Have Wait, have you tried it [44:54] yet? [44:55] >> I have not. So, my philosophy to AI is I [44:58] don't try anything until Adam tries it [45:00] first. So, um [45:04] >> No, I'm like, we talked about this [45:06] before is like I'm I'm kind of waiting [45:08] for all the tools to just figure it out [45:09] for me. like uh Claude today was like, [45:12] "Oh, you can just do stuff in the cloud [45:14] and like talk to it on your phone or [45:17] whatever." Um, so like I'm I'm waiting [45:19] for them to all figure it out. I want to [45:21] just keep blindly typing into my textbox [45:23] and like, you know, have them figure it [45:25] out. So that like Adam tries all the [45:28] more complex tools for me first and then [45:30] if it's actually a good improvement, [45:31] then I'll try it. So no, I haven't [45:33] tried. [45:35] >> That's genius. [45:35] >> I I like your description. Your [45:37] description is good. There was some [45:38] things that I liked that uh you didn't [45:41] mention and I let me bring them up just [45:44] for funsies. Um cuz you are bundling [45:46] everything in there. There's a browser [45:47] in there. Um you know like cursor just [45:50] got a visual a way to see and you're [45:53] like yeah we we already have that. Uh [45:55] I've very much appreciated you have [45:56] agents pre-built so I don't have to [45:58] define my own various set of agents that [46:00] are in the workspace. So I just assign [46:02] the task to an agent. Right now in most [46:05] of my workflows I assign a task to super [46:08] agent. I'm like I don't know pi you do [46:10] everything so go ahead and just do it or [46:13] claude you know just like here you are [46:14] the one true agent from but that's also [46:16] because I tried sub agents before that [46:18] had very defined roles never really got [46:21] anything that felt very specific out of [46:22] them but just today I tasked a design [46:25] agent to um apply a certain design [46:28] treatment to my project uh through [46:31] intent and it was very nice to see. You [46:33] also have a lot of uh really important [46:35] kind of you have work tree integrations [46:37] like just sort of like getting [46:39] everything that normally I had around me [46:42] in my workspace. Oh, holy [ __ ] We're [46:45] coming right back to everything all [46:47] bundled together or alleart you have [46:49] chosen the SAS product here. This is [46:51] intent is bundling all of the things [46:53] that you had desperately between all of [46:56] your windows and amongst many tasks. [46:58] >> Batteries included. [47:00] >> Batteries included. intent does. It's [47:02] like cursor plus Chrome plus Git plus uh [47:06] Ghosty, right? You got like and even [47:09] more than that. It's uh and so yeah, [47:10] you're kind of taking and you've called [47:12] it a workspace, but that's that's how [47:14] you're unifying those things because you [47:15] can now jump amongst the workspaces, [47:18] which is also hard on your computer, [47:20] right? I use a lot of spaces. So, I've [47:21] got I You're so funny. You're like you [47:23] have 10 or 12 tabs in a tiled window. [47:25] I'm like I I'm over here crying cuz I'm [47:27] like I literally have that right now on [47:28] my screen. And then but if I switch [47:30] spaces, you know, I swipe between my [47:31] screen, I'm like, "Oh [ __ ] there's [47:33] another eight. Oh, delightful." Uh and [47:36] now in in a workspace scenario, I can [47:38] just change the space. And so it's like [47:39] one window can facilitate all of it. So [47:41] it does seem like it's a um it's really [47:44] trying to tackle my problems as a [47:46] developer. Feels very developer [47:47] oriented. How are you coming up with [47:49] these ideas? How big is the team? Um [47:52] yeah, tell us a little bit more about [47:53] like how did you get here? because it's [47:54] also very well designed which is not [47:57] easy even with AI these days. AI does [48:00] doesn't vomit pretty [ __ ] It can vomit [48:02] terrible [ __ ] constantly. So someone is [48:05] here tending to the forest and the [48:07] trees. Um it looks really nice. [48:10] >> Oh thanks Adam. I think especially with [48:11] AI right like it becomes harder to [48:14] design when you can just like throw an [48:16] agent at it and then it like has [48:18] something and you're like I don't know [48:20] whatever [48:21] ship it. It's like semi. Okay. Um Okay. [48:25] So, the team has flexed quite a bit. Um [48:29] it's a little bit of like a glacier [48:31] product, iceberg projects where like 90% [48:36] of the work was just like figuring out [48:38] what is the shape of it, right? Like I [48:41] like [48:43] you have all these things. You have like [48:45] spectrum and development. You have like [48:47] agent collaboration and orchestration. [48:49] you have like workspaces and like that [48:52] could look like many many different [48:54] things, right? It could be a CLI tool. [48:56] It could be part of VS Code. Um, we [48:58] ended up doing like what if we didn't [49:00] use IDEs anymore because it's been [49:02] pretty restrictive trying to design [49:04] within the IDE because like all of the [49:06] space is accounted for because like it's [49:09] been like crafted over half a century to [49:12] like make it really easy to write code [49:15] in files. Um, so it actually started as [49:19] part of their VS Code extension, right? [49:21] Like what if we just had workspaces [49:23] within VS Code and then it was like this [49:25] feels pretty restricting, so let's make [49:27] it a web app. Um, and then at some point [49:29] there was a browser app. So it's like [49:32] taken many different forms. Um, and it [49:35] finally got into a shape where it was [49:37] like this kind of works and like it's [49:40] starting to sing really well. Um, and so [49:44] we launched [49:46] uh I don't know two weeks ago. Um, and [49:50] we've got like some engineers at Sutter [49:52] Hill, some at augment, and um, [49:56] it it's just been like a fun time. Like, [49:59] uh, there are people who just randomly [50:01] will give us PRs, right? Like, we want [50:04] Windows, so we'll get a PR like making [50:06] it work on Windows. And like it it's [50:09] like born out of a motivation to get it [50:11] work for working for your own workflow. [50:13] So someone's like, I'm on Linux and like [50:16] I need to like figure out how to make [50:17] intent work for me. So it's kind of like [50:19] a very ad hoc uh team right now. Um but [50:25] I'm just rambling. It's been fun. [50:29] >> Oh man, getting everything to work on [50:32] every platform is so hard for no reason. [50:35] Like [50:36] >> I was I've had a lot of woes with this [50:38] with I have a little like color picker [50:40] app called Swatch and I've just been [50:42] trying to like okay if you're on Linux [50:44] it's like cool uh that's like you got [50:47] all these I'm forgetting the terminology [50:48] now but you have like all of these [50:49] different uh like window managers and [50:51] then all of these different like uh I [50:55] want to say compositor but that doesn't [50:56] sound right. I don't know. There's all [50:58] these different like flavors of stuff [51:00] that you shouldn't have to account for. [51:02] >> Yeah. and you do and [51:04] >> yeah so like Linux support is a labor of [51:07] love for anyone that wants to do that [51:09] and like that's that's great that people [51:11] are doing that [51:12] >> at least with Linux like every all of [51:14] the users kind of want to which is kind [51:16] of the whole point right like I remember [51:19] when I used Linux and I had like an i3 [51:21] like tiling manager and you have your [51:23] own key bindings and you get like really [51:25] excited about it and then at some point [51:26] you burn out cuz it's like a lot of work [51:30] but it's fun while it lasts Um, but like [51:33] these things are so much easier than [51:36] they used to be, right? Like [51:39] imagining building a desktop app, right? [51:43] Like from scratch, especially like, oh, [51:46] what if like we rebuilt the way we use [51:50] like that sounds crazy, right? Like [51:53] that's not something that I would have [51:54] undertook a year or two ago. [51:58] >> Tell that to Zed. [52:01] Well, so they were they were ahead of [52:03] the wave. They did it on hard mode. [52:06] >> They did. [52:07] >> Yeah. Yeah. So, I guess related like, [52:10] you know, I I do think I'm spending less [52:12] and less time actually typing code and [52:14] any sort of IDE. Do you think typing in [52:17] general is just going to be dead before [52:20] we know it? Like, are we just going to [52:21] go, "Hey, uh, build me this thing." [52:23] Like, use your voice and just go, "Cool. [52:25] I've got this idea. I'm never going to [52:26] type. I don't even need a keyboard. I [52:28] just need a little microphone. Uh, do [52:30] you think we're going to get there? [52:33] >> I think it's going to be way more [52:35] prevalent, especially as things get [52:37] faster, right? Like right now, I'll do [52:40] I'll use my microphone if I have like a [52:42] lot to say, but like I'm not going to [52:45] use it for like like do that, right? Way [52:48] to talk to agents. Um, but like if you [52:53] said something and as you were talking [52:55] it updated, [52:57] there's like a speed at which I can [52:59] write with a pen, which is slower than [53:01] the speed at which I can type on a [53:02] computer, which is slower than the speed [53:04] at which I can talk with my voice, which [53:07] is slower than the speed at which I can [53:08] think. Scarily. [53:10] >> Yeah. [53:12] >> Elon. I think that's like the nice [53:14] built-in feature of typing for me is [53:16] that it's it can match my thinking [53:19] speed. If I'm talking, I could [53:21] theoretically outpace the my thinking [53:24] speed and that I don't like that. So, I [53:26] like the typing just for the feel of [53:28] like it it matches my speed. But then [53:31] like do you does that give you anxiety [53:33] during like conversations or is it just [53:35] like like I have a hotkey on my keyboard [53:38] and when it's down I'm like I really got [53:40] to get this out otherwise like you know [53:42] I'm burning tokens or whatever but like [53:45] maybe that's like not a problem in the [53:47] future. [53:49] >> Yeah. I don't know. I mean, I do feel [53:50] like I have anxiety during [53:52] conversations. Like, um, I feel like my [53:55] brain doesn't work as much anymore since [53:57] like the whole pandemic happened and [54:00] stuff. Like, I don't know. It just feels [54:01] different than it used to. I don't know [54:02] if I'm getting old or what, but yeah, I [54:04] do like I think it it's it's a feature [54:06] that that things are slower sometimes. [54:10] >> There's a phrase that's go slow to go [54:12] fast. Uh Andy Bell loves it, which is uh [54:15] being forced to slow down can actually [54:17] increase your velocity in alternative [54:20] ways. I remember that when I had my [54:21] first kid, it was right when a tool I [54:24] made came out, Visbug, and I was really [54:27] excited about it and uh couldn't stop [54:29] thinking about it. And all I wanted to [54:30] do was code on it. And I thought it was [54:32] going to be catastrophic that I had a [54:34] child in my arms that I couldn't do this [54:35] work. And I found that sitting in a [54:38] chair, uh this is also why I like [54:40] cigarettes, but that's a different [54:41] conversation. that slowing down and [54:43] spending time to think about things can [54:45] drastically improve your next moves. And [54:48] so, um, I don't know, there's some there [54:51] there's something to both things, right? [54:52] You can't you can't always move a [54:54] million miles an hour and sometimes [54:56] slowing down and thinking about the [54:57] problem space and discovering things and [54:59] leaving time for exploration. Like I [55:01] know Ameilia likes to explore. You can't [55:03] explore at the speed of light. You can, [55:05] but it doesn't give you enough time to [55:07] like soak in something that needed to [55:10] saturate. needed to marinate. Um, you [55:12] know, like a lot of the best things in [55:14] this world do happen slow, but we're so [55:16] stuck on sometimes moving really fast. [55:19] Um, I don't know. [55:21] >> Yeah, I like that. There's this anecdote [55:22] that I really like that's like semi- [55:24] true, which is like if you watch a [55:27] painter, right, like they'll they'll um [55:30] there's this loop that you go through [55:32] that everyone goes through as they build [55:34] anything, but it's like more obvious if [55:36] you look at a painter where like they [55:38] step in and they like [55:40] paint and then they like pause and they [55:43] step back and they look at it and they [55:44] like think and then they plan and then [55:47] they do the next brush stroke and there [55:49] it's like a very obvious loop of like [55:52] doing evaluating and then thinking and [55:56] like I wonder as the pacing of our work [56:00] changes like how does that loop change? [56:03] Does it get longer? Does it get shorter? [56:06] How do we make space for like the [56:08] thinking part when it's um newly easy to [56:12] do the building part? Um that's really [56:15] interesting. [56:18] >> That just blew my mind. How does how [56:20] does our loop change? Because that [56:21] that's what my stress is. That's the [56:22] that's the malth that I'm having. This [56:25] is the anxiety that I have right now is [56:26] my loop is my loop is changing and I [56:29] don't know how to deal with the new [56:30] loop. the the old loop uh I was getting [56:33] comfortable with and there's a new loop [56:36] and I don't I don't know where it ends. [56:38] I know that it feels like it's winding [56:39] faster. Um [56:42] that's a I love that. I'm going to be [56:44] chewing on that for the next week. [56:45] Ameilia, nice. [56:46] >> Maybe my next feature I ship is like uh [56:49] forced off times. [56:52] So like every you know every other five [56:55] minutes you can't do anything. [56:59] agents only [57:00] >> force Pomodoro. [57:01] >> Yeah. [57:02] >> Hey, wait. If I if an agent puts have a [57:04] cigarette on my uh canban board, that [57:07] would be kind of awesome. Be like, [57:08] "We're thinking of you, you meat bag. [57:09] You're going to need some time." [57:12] >> But I only smoke when I travel. It's not [57:14] healthy for me. So anyway, just saying [57:16] that like but I used to solve I used to [57:17] solve bugs over a cigarette all the time [57:19] where I would be there for like an hour [57:20] trying to solve something and I'd be [57:21] like, "All right, I'm just going to go [57:22] think it out." And just halfway through [57:24] and I wasn't even trying to think about [57:25] the bug. [57:26] >> I would just solve it from downtime. and [57:28] we are stealing um downtime from [57:30] ourselves at the pace that we're trying [57:32] to force our stuff into right now. [57:35] >> It's funny because I also think about [57:37] like people ask me, oh, what do you [57:39] think of like the AI coding space [57:42] >> because it's exhausting, right? Like [57:44] there's a new tool every day. There's [57:46] five every single day. Um, and like it [57:50] it's been like this since 2022 at least [57:54] like and like getting faster and faster [57:57] and like the day-to-day of it is [57:59] exhausting. But also if you step back [58:01] and you think about like what is the [58:03] actual progress of how things have [58:05] changed in 3 4 years, [58:10] it hasn't changed that much, right? Like [58:13] it's it feels like we're like kind of [58:15] running in place a little bit. Like it's [58:17] like this frantic feeling of like so [58:19] much is happening but then like on the [58:22] like meta level, right? It's like not [58:25] actually changing that much. [58:28] >> Yeah. We're 24 months into six months [58:30] until every engineer is replaced right [58:32] now. [58:34] >> Says the CEO of Hot Pockets. Soon all [58:37] dinners will be Hot Pockets, says CEO of [58:39] Hot Pockets. [58:43] >> Yeah. Um, something random I did want to [58:47] ask you about that I saw on your [58:48] website. How do I get an undo redo [58:50] spinning knob on my keyboard? Very, [58:52] >> dude. I have one, too. [58:54] >> Do you really? [58:55] >> I did not know this. [58:56] >> What do you use it for? [58:59] >> Well, undo and redo. Well, here I'll [59:00] show you. This is my [59:02] >> There's actually some warp keys on [59:04] there. Hey, warp placement. Product [59:06] placement. Um, but each of these keys. [59:08] So, yeah, here's the dials. One is [59:10] volume, obviously. Uh, that's a easy [59:12] win. But the other one is undo redo. And [59:14] so this is just a a little board. I got [59:17] to pick what all the buttons did. [59:19] >> You plug it in your computer and I [59:21] programmed it to be uh undo redo. And [59:23] that comes in really handy especially [59:25] for editing code. Um it whips, you know, [59:28] it feels really good. You can also [59:29] change its mappings per um application. [59:32] So if you wanted to make it change time [59:34] in a video editing tool or whatever, um [59:37] I like these little um customizable [59:39] keypads. They're cool. Do you um Is it [59:42] tactile or is it What do they call it? [59:45] No. What do they call it when it's um [59:48] >> Oh, haptic. It has little um bumpies as [59:51] I twist it. [59:52] >> It has bumpies. [59:53] >> It has little bumpies. Yeah, the [59:54] technical term. Little bumpies. [59:56] >> That's what I called a fart when I was a [59:58] kid. It was bumpy. [60:00] >> Bumpy fart. [60:01] >> I know. [60:02] >> Sorry. Fart joke. My bad. [60:04] >> Oh god. [60:05] >> What do you use the other buttons for? [60:08] Uh, some of them, one of them's get [60:09] pull, one of them's get push, change [60:11] spaces. So, left change space, right? [60:14] Because I was describing earlier, I [60:15] change a lot of spaces. [60:17] >> Um, one of them is expose. [60:19] >> Um, [60:21] crap, I don't remember the other ones. [60:23] >> I love how you're having to like hold it [60:24] and map to your brain of like what are [60:26] these two? [60:26] >> Yeah, it's like an instrument, right? [60:28] Yeah. I don't remember them as much as I [60:30] um my fingers do. [60:33] >> Yeah, that's cool. I'm I'm big on this, [60:35] right? Like how have we all consolidated [60:37] into the form factor of like scream [60:40] keyboard? These are the buttons everyone [60:42] has, right? Like there's a lot that [60:44] could happen here. Like I used to have [60:46] like a Wacom tablet that had [60:48] >> inspired me too. [60:50] >> Yeah. It had like a knob and it has like [60:52] buttons that you can program and like [60:55] it's really nice to be able to have [60:57] different affordances for different [60:59] activities. Uh not just have everything [61:01] work for everything. Um, but you you are [61:04] way more successful in using your [61:06] keyboard than I was on mine. [61:09] >> I I'd say that volume and the knobs are [61:12] what I primarily use. I still have [61:14] hotkey. So, git push and get pull are [61:16] still hotkey memories also like command [61:19] shiftp and GitHub desktop. Um, so I [61:22] don't always use those, but they're [61:23] there just in case. Uh, the switching of [61:25] spaces I use if I have a key. Anyway, we [61:28] could we're not talking about my [61:29] keyboard setup. Um, but yeah, it it has [61:32] been fun. Um, and I bought a small one. [61:33] There's like big ones, you know, and [61:34] some people get the especially when [61:36] they're doing live streaming, they can [61:38] switch between cameras and, you know, [61:40] and then they can be like, "All right, [61:41] you look at me from this angle now." Oo, [61:43] that's interesting, you know. [61:45] >> Yeah, we're not that fancy. We can't get [61:47] one camera to work, so I don't know if [61:48] we're [61:50] >> Don't look at our lips. First our face. [61:52] It's uh perfect. [61:53] >> Don't do it. [61:59] Okay. Yeah, we are over time. Uh, is [62:02] there anything you would like to plug or [62:04] mention before we end? [62:06] >> Um, [62:09] I did not do a good job of talking about [62:11] it, but we did just launch product. It [62:13] is free and uh we pushed very hard to [62:16] get bring your own agent. Um, augment [62:19] code has their own agent which works [62:21] really well in the app, but you don't [62:23] have to use it. You can use cloud code, [62:24] you can use codeex, you can use open [62:27] code. Um, and like we just would love [62:31] you all to try it out because it's it's [62:33] like things are changing and we want to [62:35] make it a little bit more sensical. Um, [62:39] and we'd love to have everyone along for [62:42] the ride to just like try it out with us [62:44] and iterate on, you know, what happens [62:46] when [62:48] agents do most of the code. What does [62:50] our lives look like? [62:53] >> We'll put the link to all your products [62:55] in the show notes cuz there are more. [62:57] Yeah, we only touched on intent and um [63:00] y'all have other offerings, so we'll [63:02] have to see uh what those are in our [63:05] show notes, everybody. [63:07] >> Keep it dramatic. [63:08] >> Or follow Amelia, who's going to have [63:10] all of these things on their profile for [63:11] the next little bit, I'm sure. [63:13] >> I'm sure we will talk about them in [63:15] future episodes after having tried all [63:17] the things because that's all we talk [63:18] about now is AI. So stay tuned for an [63:21] update sometime in the future. [63:26] All right. Uh, thanks everyone for [63:27] listening. If you liked it, please [63:28] subscribe. Leave us some ratings and [63:30] reviews. We appreciate it. And we will [63:31] catch you next time.  [[3. Resources/Readwise/Full Document Contents/Articles/The Manager Has Become The Managed|Full Document Contents]] ## Document Notes --- tags: #ai-development #developer-experience #automation #software-engineering author: Adam Argyle source: YouTube reference: https://www.youtube.com/watch?v=Zja5HO4Ztp4 date created: 2024-12-19 12:00 --- # 🏷️ Title: The Manager Has Become The Managed ## 🧰 Action Inbox - [H1] Explore AI agent orchestration tools for managing code review bottlenecks — addresses role reversal from manager to managed #automation - [H2] Implement spec-driven development workflows to maintain control over AI output — prevents overwhelming PR creation rates #software-engineering - [H3] Evaluate workspace-based development environments that bundle all tools — reduces context switching between terminals/apps #developer-experience ## 🧭 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 - [Intent - Workspace-based Development Tool](URL) - [Augment Code](URL) - [OpenClaw](URL) ## 📝 Summary - Developers are experiencing a role reversal where AI agents create more work than humans can review, making managers feel "managed" by their tools - AI chatbots may not be the optimal interface for development workflows; specialized tools with embedded context work better - Traditional development craft is evolving from writing code to orchestrating AI agents and systems - The backlog has become valuable again as AI can quickly implement previously stalled ideas - New workspace-based development environments bundle all necessary tools to reduce context switching ## 📖 Main Content The conversation explores how AI is fundamentally changing the developer experience, with a focus on the unexpected reversal where developers find themselves overwhelmed by AI-generated work rather than empowered by it. The speakers discuss how traditional chatbot interfaces are limiting and explore emerging solutions like workspace-based development environments. Key themes include the evolution from individual coding to agent orchestration, the renewed importance of planning and specification, and how the craft of programming is shifting toward higher-level system design and management. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/Zja5HO4Ztp4?t=0) — [H1] Introduction and whiskey tasting setup - Key idea: Setting up discussion about AI development tools with guest Amelia Watenberger - Evidence: "This is Amelia Watenberger, digital creatives, products spelunker, codebase revealer, data visualizer" #### ⏱️ [12:30](https://youtu.be/Zja5HO4Ztp4?t=750) — [H2] AI chatbots as limitations rather than solutions - Key idea: Chatbots are jack-of-all-trades but master of none - Evidence: "chatbots are kind of the terminal of the next phase... they do everything pretty well but they don't do anything really well" #### ⏱️ [18:45](https://youtu.be/Zja5HO4Ztp4?t=1125) — [H1] The manager has become the managed phenomenon - Key idea: AI agents now create more tasks than developers can handle - Evidence: "a bot started creating tasks for me... all of a sudden I am now the other end of the vibe kanban" #### ⏱️ [25:20](https://youtu.be/Zja5HO4Ztp4?t=1520) — [H2] Loss of programming craft and satisfaction - Key idea: Fast AI solutions eliminate the struggle and victory cycle of debugging - Evidence: "I used to run, I used to get up at office all the time... now it happens with AI in like one prompt" #### ⏱️ [32:10](https://youtu.be/Zja5HO4Ztp4?t=1930) — [H3] Backlog renaissance in the AI era - Key idea: Previously dead backlogs are now actionable with AI implementation - Evidence: "Joel Hooks was saying that our backlogs cool again... Your backlog used to be a dead weight" #### ⏱️ [38:50](https://youtu.be/Zja5HO4Ztp4?t=2330) — [H2] Burnout and new types of stress with AI - Key idea: AI doesn't reduce stress but creates different kinds of unhealthy relationships with computers - Evidence: "I'm in a new type of unhealthy relationship with my computer... I'm not better. I'm not doing less" #### ⏱️ [45:15](https://youtu.be/Zja5HO4Ztp4?t=2715) — [H3] Interface similarity problems across AI tools - Key idea: Similar chat interfaces cause confusion between different AI agents - Evidence: "I find myself... I'm not always very polite to my agents... I've almost typed agent things into Slack" #### ⏱️ [52:30](https://youtu.be/Zja5HO4Ztp4?t=3150) — [H1] Voice-to-visual ideation experiments - Key idea: Exploring new mediums beyond text chat for AI interaction - Evidence: "transcribed as you were talking and then turned your concepts into index cards" #### ⏱️ [58:20](https://youtu.be/Zja5HO4Ztp4?t=3500) — [H2] Data liberation and personalized applications - Key idea: Moving from siloed apps to personal data control and custom experiences - Evidence: "I'm liberating my data... when it's mine, I can build my own app with it" #### ⏱️ [65:40](https://youtu.be/Zja5HO4Ztp4?t=3940) — [H1] Intent workspace-based development introduction - Key idea: Bundling all development tools into unified workspaces - Evidence: "if we put everything to do with a task into like a bag... we get like the crazy orchestration workflows" #### ⏱️ [72:15](https://youtu.be/Zja5HO4Ztp4?t=4335) — [H2] Spec-driven development with AI agents - Key idea: Using specifications to control and organize AI-generated code - Evidence: "I write down what I want to do. An agent goes and explores my codebase and writes a spec for me" #### ⏱️ [78:50](https://youtu.be/Zja5HO4Ztp4?t=4730) — [H3] Evolution from IDE-based to workspace-based development - Key idea: Moving beyond traditional IDEs that were designed for manual coding - Evidence: "what if we didn't use IDEs anymore because it's been pretty restrictive trying to design within the IDE" ## 📚 References - [Intent Desktop App](https://intent.dev) - [Augment Code](https://augmentcode.com) - [Fullstack D3 and Data Visualization](https://www.newline.co/fullstack-d3) - [GitHub Next Team](https://githubnext.com) ## 🗣️ Paraphrase - AI development tools are creating a role reversal where developers feel managed by their agents rather than managing them, as bots generate more work than humans can review - Traditional chatbot interfaces are insufficient for complex development workflows, leading to exploration of new interaction paradigms like voice-to-visual ideation and workspace-based environments - The craft of programming is evolving from manual coding to orchestrating AI systems, requiring more upfront planning and specification while potentially eliminating the satisfying struggle-victory cycle of debugging ## 📰 Newsletter The future of developer experience is being reshaped by AI, but not in the ways we expected. Instead of making developers more productive, many are finding themselves overwhelmed by AI-generated code and pull requests they can't review fast enough. This "manager becomes managed" phenomenon highlights a critical need for better orchestration tools and interfaces beyond simple chatbots. New approaches like workspace-based development environments are emerging to bundle all necessary tools and context, while spec-driven development helps maintain human control over AI output. The craft is evolving from writing code to designing systems and managing AI agents. Resources: [Intent Workspace Tool](https://intent.dev), [AI Development Best Practices](https://developer.wordpress.org/), [Agent Orchestration Patterns](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer advocacy: Create content about managing AI agent workflows, present at conferences on "post-chatbot" development experiences - Education: Develop workshops on spec-driven developme ## Highlights - and it goes uh good human and puts this like cute little set of emojis and I was like I I well and then that's what I mean. I'm like I'm feeling managed that my computers are managing me. Uh anyway, have y'all experienced this at all? you is this is this the future that we're headed towards? Is we're going to be surrounded by more agents than we can think of or is there a tool uh or a way that we can stay in control of this wild amount of output? ([View Highlight](https://read.readwise.io/read/01kjb9smcqhedsesmde2n5yhwm)) - Tags: 👻 ai highlighted - feels to me. >> Yeah, I've been experiencing this a lot of like I had so many ideas, so many things to work on. Every time I had any free time, I'm jumping on my like personal projects like you know, hacking away and now I'm like it's kind of done. Like I don't I don't know what else I can do. Like it feels weird. Um, so yeah, you have to be like, it's almost like the focus on, you know, hacking and programming is just totally gone. And ([View Highlight](https://read.readwise.io/read/01kjb9smf0gjeceb2ad59hcz7k)) - Tags: 👻 ai highlighted - that I'm coding. I'm coding an entire operating system to do functionality for me that's much more grandiose than like an agent was doing. And so we are we're changing our craft. We're adapting and we're solving bigger better problems maybe um because of it. And it it is exciting. I'm I'm finding a lot of it thrilling in a new way. I'm just some of it is kind of funny. the realities that are coming back. Like I am I am literally getting told what to do by agents sometimes and I'm like I don't want to review that. Okay, whatever. I will. Um but yeah, ([View Highlight](https://read.readwise.io/read/01kjb9smp9j2kf71qx4jhza0pz)) - Tags: 👻 ai highlighted - >> Um so yeah, I mean there's a lot of things like how do you fix that problem of saying I I don't want you to compact that. I just want you to always remember my instructions. Yeah, that's kind of the fun thing about building tools out of these agents is like figuring out the new parts that are rigid and the new parts that are flexible. Um cuz it feels like even like that paradigm of an app like doesn't fit anymore and like we kind of have to figure out ([View Highlight](https://read.readwise.io/read/01kjb9smt3t9ybzs0pzbf28smg)) - Tags: 👻 ai highlighted - first. So, um >> No, I'm like, we talked about this before is like I'm I'm kind of waiting for all the tools to just figure it out for me. like uh Claude today was like, "Oh, you can just do stuff in the cloud and like talk to it on your phone or whatever." Um, so like I'm I'm waiting for them to all figure it out. I want to just keep blindly typing into my textbox and like, you know, have them figure it out. So that like Adam tries all the more complex tools for me first and then ([View Highlight](https://read.readwise.io/read/01kjb9sn0zv5n7qs26yc63d8hy)) - Tags: 👻 ai highlighted - spaces, you know, I swipe between my screen, I'm like, "Oh [ __ ] there's another eight. Oh, delightful." Uh and now in in a workspace scenario, I can just change the space. And so it's like one window can facilitate all of it. So it does seem like it's a um it's really trying to tackle my problems as a developer. Feels very developer oriented. How are you coming up with these ideas? How big is the team? Um yeah, tell us a little bit more about like how did you get here? because it's also very well designed which is not easy even with AI these days. AI does doesn't vomit pretty [ __ ] It can vomit ([View Highlight](https://read.readwise.io/read/01kjb9snf27w5qhwtw1mq59zr9)) - Tags: 👻 ai highlighted - but it's fun while it lasts Um, but like these things are so much easier than they used to be, right? Like imagining building a desktop app, right? Like from scratch, especially like, oh, what if like we rebuilt the way we use like that sounds crazy, right? Like that's not something that I would have undertook a year or two ago. >> Tell that to Zed. ([View Highlight](https://read.readwise.io/read/01kjb9snpywpyvtzxzfttz8fyf)) - Tags: 👻 ai highlighted - ourselves at the pace that we're trying to force our stuff into right now. >> It's funny because I also think about like people ask me, oh, what do you think of like the AI coding space >> because it's exhausting, right? Like there's a new tool every day. There's five every single day. Um, and like it it's been like this since 2022 at least like and like getting faster and faster and like the day-to-day of it is exhausting. But also if you step back ([View Highlight](https://read.readwise.io/read/01kjb9sp90nw9qffhpavg2wgy0)) - Tags: 👻 ai highlighted - um my fingers do. >> Yeah, that's cool. I'm I'm big on this, right? Like how have we all consolidated into the form factor of like scream keyboard? These are the buttons everyone has, right? Like there's a lot that could happen here. Like I used to have like a Wacom tablet that had >> inspired me too. >> Yeah. It had like a knob and it has like buttons that you can program and like it's really nice to be able to have different affordances for different activities. Uh not just have everything ([View Highlight](https://read.readwise.io/read/01kjb9spr5kq546f0mn5cq94mq)) - Tags: 👻 ai highlighted - you all to try it out because it's it's like things are changing and we want to make it a little bit more sensical. Um, and we'd love to have everyone along for the ride to just like try it out with us and iterate on, you know, what happens when agents do most of the code. What does our lives look like? >> We'll put the link to all your products in the show notes cuz there are more. Yeah, we only touched on intent and um ([View Highlight](https://read.readwise.io/read/01kjb9sq5xy6vq95j3esq3m421)) - Tags: 👻 ai highlighted