We Accidentally Deleted Programming (Presented by CodeRabbit) [01kh207nwrktmavgyqsv2yqn0r]

We Accidentally Deleted Programming (Presented by CodeRabbit) 🔗 **Source:** [We Accidentally Deleted Programming (Presented by CodeRabbit)](https://www.youtube.com/watch?v=9w_5Pstg1zk) > [!note]+ Document Notes > tags: [ai-development, code-review, javascript-frameworks, whiskey-review] > author: Whiskey Web and Whatnot > source: YouTube > reference: https://www.youtube.com/watch?v=9w_5Pstg1zk > date created: 2024-12-19 10:30 > > # 🏷️ Title: We Accidentally Deleted Programming (Presented by CodeRabbit) >  ## 🧰 Action Inbox - [H1] Try CodeRabbit for AI code review — 100k open-source projects using it #ai-tools - [H2] Experiment with anchor positioning for FLIP animations — now baseline in Firefox #css - [H3] Evaluate library conversion from Svelte 5 to Solid — bundle size concerns #javascript ## 🧭 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 - [CodeRabbit](https://coderabbit.ai/) - [Whiskey Web and Whatnot Podcast](https://whiskeywebandwhatnot.fm/) ## 📝 Summary - Robbie Wagner and Adam Argyle discuss the impact of AI on programming while reviewing a Whistle Pig Liquid Death collaboration whiskey - They explore whether traditional coding practices like TypeScript and databases still matter in an AI-first development world - Discussion covers new CSS features like anchor positioning, framework bundle sizes, and the evolution of developer tools - The conversation highlights the tension between maintaining code quality and embracing AI-generated solutions - Bottom line: Programming is evolving from writing code to writing specifications and prompts for AI systems ## 📖 Main Content The hosts examine how AI is fundamentally changing software development, questioning whether traditional practices remain relevant. They discuss the emergence of "spec-driven" libraries that ship instructions for LLMs rather than actual code, the challenges of maintaining TypeScript in large codebases when AI can provide similar guardrails, and new CSS features that eliminate the need for JavaScript solutions. The conversation reveals a industry in transition, where the ability to effectively prompt AI may become more valuable than deep coding knowledge. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/9w_5Pstg1zk?t=0) — [H1] Whiskey Review: Whistle Pig Liquid Death Collaboration - Key idea: 85% wheat, 15% barley whiskey aged in coffin-shaped barrels - Evidence: "86 proof, proofed with liquid death mountain water, aged final months in 380 gallon charred American oak casket" #### ⏱️ [05:23](https://youtu.be/9w_5Pstg1zk?t=323) — [H2] AI Automation for Content Creation - Key idea: Building Swift app to automate podcast episode artwork generation - Evidence: "Pull title, show notes, transcripts, pull 5-10 topics, make them concrete things you could make an image of" #### ⏱️ [08:45](https://youtu.be/9w_5Pstg1zk?t=525) — [H1] CSS Anchor Positioning Replaces JavaScript FLIP - Key idea: New CSS features can recreate complex JavaScript animations natively - Evidence: "Anchor positioning, interpolate-size, and starting-style can do FLIP animations in pure CSS" #### ⏱️ [12:30](https://youtu.be/9w_5Pstg1zk?t=750) — [H2] Svelte 5 Bundle Size Issues - Key idea: Framework upgrades don't always improve bundle sizes for small component libraries - Evidence: "Went from 17 to 28 kilobytes, core is bigger but components are smaller, doesn't benefit with only 5 components" #### ⏱️ [15:15](https://youtu.be/9w_5Pstg1zk?t=915) — [H1] Libraries That Ship No Code - Key idea: New paradigm of shipping specs and instructions for LLMs instead of actual code - Evidence: "Ships the specs and expectations so that an LLM can do it in any language, works across Elixir, Ruby, Rust" #### ⏱️ [18:20](https://youtu.be/9w_5Pstg1zk?t=1100) — [H2] AI-First Hiring Practices - Key idea: Companies moving away from coding interviews toward prompting skills - Evidence: "Show me how you can prompt and let's see how productive you can be, I don't care how you do it" #### ⏱️ [22:10](https://youtu.be/9w_5Pstg1zk?t=1330) — [H1] TypeScript Relevance in AI Era - Key idea: Questioning whether strict typing matters when AI provides similar guardrails - Evidence: "What if you spent time investing in guardrails for AI instead of converting to TypeScript?" #### ⏱️ [28:30](https://youtu.be/9w_5Pstg1zk?t=1710) — [H2] AI Struggles with Large Codebases - Key idea: Current AI models still have significant limitations in complex, multi-file projects - Evidence: "Past four files, it starts to go downhill, even CodeQX 5.2 specialized for long-running tasks" #### ⏱️ [32:45](https://youtu.be/9w_5Pstg1zk?t=1965) — [H1] Developer Tool Evolution - Key idea: Modern AI-powered development environments offering multiple approaches - Evidence: "Cursor has agent modes, work trees, multiple agents working in parallel with visual diff comparison" #### ⏱️ [36:20](https://youtu.be/9w_5Pstg1zk?t=2180) — [H2] Multi-Model AI Development - Key idea: Using different AI models competitively for better results - Evidence: "Polyamorous with models, pit Gemini against Claude against ChatGPT, have them review each other's work" #### ⏱️ [40:15](https://youtu.be/9w_5Pstg1zk?t=2415) — [H1] MCP Server Integration - Key idea: AI tools automatically configuring and discovering the right plugins for tasks - Evidence: "MCP discovery where I ask for git stuff and it pulls GitKraken MCP to figure out how git works" #### ⏱️ [43:50](https://youtu.be/9w_5Pstg1zk?t=2630) — [H2] File System Access Concerns - Key idea: Balancing AI automation with security and control over development environment - Evidence: "Co-worker making files and has file system access, not ready for automated file access yet" ## 📚 References - [FLIP Animation Technique](https://aerotwist.com/blog/flip-your-animations/) - [CSS Anchor Positioning](https://developer.mozilla.org/en-US/docs/Web/CSS/CSS_anchor_positioning) - [Svelte 5 Documentation](https://svelte.dev/) - [OpenCode AI Development Tool](https://opencode.dev/) ## 🗣️ Paraphrase - AI is fundamentally changing how we approach software development, shifting focus from writing code to writing specifications and effective prompts - Traditional development practices like strict TypeScript typing may become less relevant as AI provides similar error prevention through different mechanisms - New CSS features are eliminating the need for JavaScript solutions in areas like animations and positioning, reducing framework dependencies - Current AI tools still struggle with large, complex codebases but are rapidly improving and may eventually handle most routine development tasks ## 📰 Newsletter **The Future of Web Development: When AI Writes the Code** This week's episode explores how AI is reshaping web development fundamentals. Key highlights include new CSS anchor positioning that can replace JavaScript animation libraries, the emergence of "spec-only" libraries that work with any programming language through AI, and the ongoing debate about whether traditional practices like TypeScript still matter in an AI-first world. For WordPress developers, this shift toward AI-assisted development could dramatically change how we approach theme and plugin development, making specification writing and effective prompting increasingly valuable skills. Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer experience: Document AI-assisted development workflows for community adoption - Education: Create workshops on effective prompting for developers transitioning to AI tools - Developer marketing: Showcase AI-enhanced productivity in development campaigns #### Community - Contributors: Develop guidelines for AI-generated code contributions and review processes - Maintainers: Establish protocols for evaluating AI tools and their impact on code quality #### WordPress - Core: Evaluate impact of AI-generated contributions on codebase quality and maintenance - Documentation: Create guides for using AI tools effectively in WordPress development - Training: Develop curriculum combining traditional development skills with AI prompting techniques ## 🧪 Hypotheses - Developers who master AI prompting will be more productive than those who rely solely on traditional coding skills - CSS anchor positioning will significantly reduce JavaScript bundle sizes in modern web applications - Spec-driven development will become the dominant paradigm for cross-language library development - TypeScript adoption will slow as AI tools provide equivalent error prevention with less overhead ### 📊 Methodology - Comparative study tracking developer productivity with AI tools vs traditional coding - Bundle size analysis measuring impact of CSS anchor positioning adoption - Survey of hiring practices examining shift toward prompting skills assessment ### ❓ Questions - How do we maintain code quality standards when AI generates increasing amounts of production code? - What skills should developers prioritize learning in an AI-first development world? - How will the role of senior developers evolve when junior developers have access to advanced AI assistance? ### 🔮 Future Research - Long-term impact of AI-generated code on software maintenance and technical debt - Effectiveness of different AI model combinations in complex development tasks - Evolution of development education and certification in AI-augmented environments ### 💡 Implications - Practice: Developers need to balance AI tool adoption with maintaining core programming competencies - Education: Curriculum must evolve to include effective AI prompting alongside traditional programming concepts - Security: AI file system access and automated code generation create new attack vectors requiring consideration - Funding: Investment in AI development tools may yield higher returns than traditional development infrastructure ### 🧠 Learning Styles Assessment Reading level estimate: 10th Grade #### 👀 Visual learners - Present: Code examples, framework comparisons, bundle size visualizations - Improve: Flowcharts showing AI development workflows, visual comparisons of traditional vs AI approaches #### 🔊 Aural learners - Present: Podcast format with detailed audio discussion - Improve: Audio summaries of key technical concepts, pronunciation guides for new terminology #### 🗣️ Verbal learners - Present: Conversational format with analogies and real-world examples - Improve: Glossary of AI development terms, story-driven explanations of complex concepts #### 🧩 Kinesthetic learners - Present: Discussion of hands-on experimentation with tools - Improve: Step-by-step tutorials for trying discussed AI tools, interactive coding exercises ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear heading structure, descriptive timestamps - Improve: Alternative text for any referenced visual elements, transcription formatting #### Contrast and readability - Present: High contrast text formatting in notes - Improve: Color-coded sections for different topics or priority levels ### 🌍 DEIB Assessment #### Cultural breadth - Present: Discussion includes various development contexts and company sizes - Improve: More diverse examples of how AI impacts developers globally #### Inclusive language - Present: Gender-neutral terms, accessible technical explanations - Improve: Plain-language summaries of complex technical concepts ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Clear logical progression from AI capabilities to development practice implications - Inductive strength: Examples from real-world tool usage support broader conclusions - Analogical strength: Effective comparisons between traditional and AI-assisted development workflows #### Tone of voice - Overall tone: Conversational and exploratory, appropriate for discussing rapidly evolving technology landscape ### 🛡️ DARVO Analysis No evidence of denial, attack, or victim/offender reversal patterns. Discussion maintains balanced perspective on AI tool benefits and limitations. ## 🎯 Conclusion The conversation reveals an industry in rapid transition, where traditional programming skills are being augmented—and in some cases replaced—by AI systems. The key takeaway is that successful developers will need to evolve their skillset to include effective AI collaboration while maintaining enough foundational knowledge to evaluate and guide AI-generated solutions. The shift toward specification-driven development suggests a future where the ability to clearly communicate requirements and constraints may become more valuable than the ability to implement solutions manually. ## 🧱 Refactor Appendix ### 📚 Literature Split Pad ### [H1] AI fundamentally transforms development workflow > [!quote] [We Accidentally Deleted Programming](https://www.youtube.com/watch?v=9w_5Pstg1zk) by [[Whiskey Web and Whatnot]] via [[YouTube]]. Published 2024. Accessed 2024-12-19. > "What if you spent time investing in guardrails for AI instead of converting to TypeScript?" Represents shift from traditional code quality tools to AI-powered alternatives. [view highlight](https://youtu.be/9w_5Pstg1zk?t=1330) #ai-development #typescript ID:rw_001 ### [H2] Spec-driven libraries eliminate language barriers > [!quote] [We Accidentally Deleted Programming](https://www.youtube.com/watch?v=9w_5Pstg1zk) > "Ships specs and expectations so LLM can do it in any language, works across Elixir, Ruby, Rust." New paradigm where libraries provide instructions rather than implementations. [view highlight](https://youtu.be/9w_5Pstg1zk?t=915) #library-design #cross-platform ID:rw_002 --- ### 🔖 Atomic Split Pad (Zettel Candidates) #### [Z1] CSS anchor positioning replaces JavaScript animations - Scope: Modern browsers with baseline support - Pointer: 08:45 timestamp discussion - Links: [[CSS-Features]] [[JavaScript-Alternatives]] #### [Z2] AI prompting skills become hiring criteria - Scope: Technology companies adapting interview processes - Pointer: 18:20 timestamp discussion - Links: [[Developer-Skills]] [[AI-Integration]] --- ### ✏️ Fleeting Queue - [F1] Test CSS anchor positioning for FLIP animations — https://youtu.be/9w_5Pstg1zk?t=525 #fleeting #css-experiments - [F2] Evaluate CodeRabbit for team code review process — https://coderabbit.ai/ #fleeting #ai-tools - [F3] Research multi-model AI development approaches — https://youtu.be/9w_5Pstg1zk?t=2180 #fleeting #ai-workflow --- ## Full Transcript [00:02] Welcome to syntax. Welcome to a brand [00:04] new episode of the frontend happy hour [00:06] podcast. Welcome to this week's JS Party [00:12] >> live from Ship Shape Studios. This is [00:15] Whiskey Web and Whatnot with your hosts [00:18] Robbie the Wagner and me, Charles [00:21] William Carpenter III. [00:23] >> That's right, Charles. We drink whiskey [00:25] and talk about web development. [music] [00:27] I mean, it's all in the name. It's not [00:29] that deep. [00:30] >> This is whiskey web and whatnot. Do not [00:33] adjust your set. [00:36] >> As more and more code is generated, [00:38] developers end up spending more time [00:39] reading and reviewing code. This creates [00:41] an issue on reliably shipping quality [00:42] code. So even though AI is promising, a [00:45] lack of proper guardrails can slow down [00:46] teams. Code Rabbit is an AI code [00:48] reviewer. Unlike AI code generators like [00:51] Cursor, Windsurf or GitHub Copilot, it [00:53] reasons and submits feedback on your [00:55] codebase while you raise a pull request [00:56] to merge code into your VCS. Code Rabbit [00:59] is a self-improving agent and it learns [01:01] from your interactions and creates [01:02] learnings. Code Rabbit generates [01:04] one-click commit suggestions for some [01:06] review comments, making it easy for [01:08] developers to address review comments. [01:10] Code Rabbit is used by 100,000 open- [01:12] source projects like B, Nex.js, and it's [01:14] installed on 2 million repositories so [01:16] far. It's also performed more than 15 [01:18] million PR reviews. [01:22] >> Hey, what's up everybody? Welcome to [01:24] Whiskey Web and Whatnot with your hosts [01:27] Robbie the Wagner and Adam Thomas Argyle [01:30] the Nerd. What's going on? [01:31] >> What's up y'all? I'm feeling good. How [01:33] are you? [01:34] >> I am good as well. It is once again [01:36] Margarita Thursday, which I just call it [01:39] this to make my wife upset because it is [01:42] clearly either thirsty Thursday or [01:44] Margarita Monday, but I like to do the [01:46] dumb one that doesn't make any sense. [01:48] Oh, nice. We do uh Woody Wednesdays. We [01:50] go to Little Woodies and then we used to [01:52] do like Fridays. Well, that was really [01:55] stretchy, [laughter] you know. [01:58] >> Now it Pokemon. That one always stayed [02:01] good. It's not a Friday, but Friday [02:03] doesn't start with P. It's okay. [02:04] >> Yeah. Yeah, these things are hard. All [02:07] right, so I'm excited for the whiskey we [02:09] have today. This is a special edition. [02:12] Where's my camera at? Whistle Pig Liquid [02:15] Death collaboration. I'm a big Liquid [02:17] Death fan. I really like their uh Dr. [02:19] Pepper flavor. I forget what they call [02:21] it. Dr. Death, maybe. [02:22] >> I should try that. I love Dr. Pepper. [02:24] >> Yeah. So, is this is a wheat whiskey. It [02:27] is 85% wheat and 15% barley. They use [02:31] liquid death mountain water to proof it [02:33] down, which like hopefully they got a [02:35] discount on it because liquid death is [02:37] not [laughter] cheap. Can even they're [02:39] like popping all the cans and pouring [02:40] them in there. Um, yeah, it's 86 proof. [02:44] I don't know what the age is, but like [02:46] you get the idea. Liquid death [02:48] collaboration. Looks cool. Smells sweet. [02:52] Smells fruity. [02:55] Yummy. Okay. Yeah. I can never tell. [02:59] Like sometimes there's a little funk and [03:01] I can't tell if like my glass is dirty [03:04] or [03:05] >> it's got baby stuff in there. Yeah. [03:07] >> Yeah. [03:08] >> Shouldn't have mixed similac in my uh [03:11] nice norling glass. Why' I do that? [03:13] [laughter] [03:15] >> Yeah, I do. I feel like I smell [03:16] strawberry. Strawberry cinnamon perhaps. [03:19] I'm going to taste it. [03:20] >> Oh man, I'm getting like a berry medley. [03:23] It's nice. Really light spice. Smooth [03:26] and soft. [03:27] That is yummy. [03:29] >> Uh coating of my tongue from the front [03:32] to the back. So I got like the spice and [03:33] then just the the saturation as it [03:36] spreads across. [03:38] Yeah, this is very good. I guess maybe [03:40] some of the similar notes that I like [03:42] from rise maybe come through in this cuz [03:44] it's all like uh grains and I guess is [03:47] corn a grain? No, I guess [03:49] >> it is not. [03:50] >> Yeah. So like it doesn't have corn. I [03:51] don't know. [03:52] >> Which I feel like makes it uh tastier [03:55] like taking the corn out. [03:58] I'm glad there's still some spice. [04:00] Always like a little spicy bite. And [04:02] then uh it's pretty low on the proof. Do [04:04] you do you pay for proof? Like do are [04:06] you like, "Oh, I'm if I'm going to spend [04:08] money, it better be high proof." Do you [04:10] do you care about that? Is that what [04:12] >> No. No. I just care about [04:14] >> Yeah. [04:14] >> Yeah. I care about interesting things [04:16] and it being like if the distiller wants [04:18] me to try it at 135 proof or whatever, [04:21] like, okay, like I'm down, but I'm not [04:24] seeking that out. I I usually like it to [04:26] be a little higher than like uh 80 proof [04:29] or whatever. Like whatever the standard [04:31] is. So this is a little bit higher, I [04:32] think, right? It was Yeah. 86. So it's [04:34] got a little more burn. Um yeah, I think [04:37] it's very balanced. Like doesn't burn [04:38] too much. Nice spice, fruity notes. [04:42] Pretty good. [04:43] >> Pretty good. I'm happy. So yeah. What [04:45] What's your rating? [04:46] >> Well, I don't know what the scale. Do [04:48] you do you remember what our scale is? [04:50] >> Zero to eight. If it's zero, you're [04:52] like, throw that [ __ ] away. If it's an [04:54] eight, you're like, "Clear the shelves. [04:55] I want this forever. Just put it in my [04:57] belly." Uh, and then, you know, anything [04:59] in the middle. Um, yeah. What you got? [05:02] >> Um, yeah. I don't have anything bad to [05:05] say about it. Uh, but it also is maybe [05:08] not like my favorite one ever. It's so [05:09] hard. We've had so many. I'm just going [05:11] to say flat seven. [05:13] >> Flat seven. Solid. I was thinking six, [05:15] six and a half cuz Yeah, it's uh solid. [05:18] I like it. I will have it again. I do [05:20] like that it's kind of soft and sweet. [05:22] It has it. Maybe that's the low. It's [05:24] the low proof is probably the softness. [05:26] >> And the mountain water. [05:27] >> And the mountain water. Yeah, that that [05:29] matters. Oh. Oh, I was reading about [05:31] this, too. Um, it was uh aged in a [05:35] coffin. [05:36] >> Wait, was it really? [05:37] >> It was really So, they started out in an [05:39] American oak like regular barrel, but [05:42] they move it from a barrel into a [05:44] coffin. [05:46] >> Uh, which is why it has the uh coffin [05:49] symbol on it, and it's called gravetock. [05:51] So that's why it's called gravetock is [05:52] there's an American oak coffin. [05:54] >> Oh yeah. [05:55] >> That apparent and they're like teasing [05:56] that it's [05:57] >> final months resting in peace in a 380 [05:59] gallon charred American oak casket [06:01] [laughter] [06:02] >> then was proofed with liquid death [06:04] mountain water. Wow. [06:07] Okay. [06:09] >> Love it. I like that dedication to [06:11] detail. [06:12] >> Yeah. [06:13] Money to burn. [06:15] [laughter] [06:16] >> Oh, dude. Speaking of like death and and [06:19] American wood, here, check out this [06:20] banjo I bought myself for my 40th [06:22] birthday here. [06:23] >> It's already beat up a zombie. [06:26] [laughter] [06:28] >> Nice. Love it. [06:29] >> Covered in blood. It's covered in little [06:31] zombie icons. Uh there's like a [06:34] >> knife through his head there. There's [06:35] someone's foot. Uh it's called the [06:37] zombie killer. And it is very cool. [06:40] Here, let's see. Oh, I don't ever play [06:43] this way, so it's weird. I'm going to [06:44] [snorts] try to play Mario. [06:46] Nope. I'm not going to be able to do it [06:48] standing here like that. [laughter] [06:51] >> Hey, it sounds good. [06:53] >> It does sound good. It's great. [06:54] >> Having only heard you play banjo like [06:57] lowfi trying to throw it up against the [06:59] mic, I feel like this one sounds better [07:01] than the other one you had. [laughter] [07:03] >> Cool. I'm hoping this is my heirloom. [07:05] Like, dad's dead, but don't worry, we've [07:06] got his zombie killer uh banjo, so he'll [07:09] live on forever with that sick banjo. [07:11] It'd be cool if they like reinforced it [07:13] where the like splatters are so you [07:15] could like actually hit it on stuff and [07:17] it'd be fine. [laughter] [07:19] >> That would be cool. Did you see the saw [07:20] that was around the edge, too? So, the [07:22] ring, there's like a tone ring, but [07:24] around the ring is a saw. Like straight [07:26] up edges of a saw. It's super cool. [07:28] >> That sounds dangerous. [07:30] >> Yeah, I cut my finger. Ah, [laughter] I [07:33] should put some of my real blood on [07:34] there just cuz if it's a heirloom. [07:36] >> Oh, man. I used to bleed on my guitar so [07:38] much when I would play shows because I [07:40] would like hit hit it wrong and catch my [07:42] like nail or something and just like [07:43] >> Oh, yeah. [07:44] >> be just pouring blood all over the [07:46] guitar for like 10 minutes. [07:47] >> Cool. [07:48] >> Yeah. [laughter] [07:49] >> Yeah. Yeah. We have to find some [07:50] recordings of of you uh in some bands [07:52] cuz I I was saying we should do I don't [07:54] know if we get Chuck or somebody else to [07:56] do it, but we need people to listen to [07:58] like a little bit of band I was in and a [08:00] little bit of band you were in. See if [08:01] they can say who was who. I think that'd [08:03] be a fun game. I mean, I feel like it's [08:05] going to be kind of obvious, but I don't [08:06] know. Maybe people don't know the past [08:08] emo me and they'll be like, "Who's this [08:10] young kid singing?" [laughter] [08:13] >> Yeah. Yeah. I mean, if you if you don't [08:15] know like which uh part of the band you [08:18] were, I guess is like maybe [08:20] >> Oh, so you didn't sing. Okay. Okay. [08:22] That's cool. [08:23] >> Yeah. Well, I did do some singing. Um [08:25] but like Yeah. So, we'll see. We'll see. [08:30] We got to uh finish the developer band [08:33] as well sometime. [08:35] >> I want to [08:35] >> so many things on the stinking to-do [08:37] list. It drives me nuts. My to-do list [08:39] is so big. There's like stuff on there. [08:41] I'm like, I would like to get to that. [08:42] And I'm like, yeah, but there's stuff I [08:43] haven't done for the past 3 days that I [08:45] need to do. So, [08:47] >> yeah, [08:47] >> I'll be an adult, I guess. Um [08:51] >> Yep. That's what I unfortunately pick on [08:54] the daily, but uh maybe one day we won't [08:56] have to pick that. I like the uh the [08:58] irony that I I texted you too. It's like [09:00] we need to spend some time automating AI [09:02] to do some of our tasks so that we can [09:05] have more time. And I'm like wait that's [09:06] a chicken and egg. I'm like if I don't [09:08] have the time I can't create the [09:10] automations to save me the time. I'm [09:12] like [09:13] >> agents are great for that cuz like they [09:15] might not get it for me every time but I [09:17] can at least be like hey I'd like this [09:19] and then come back later and if it [09:20] happened to do it then I've got it. Like [09:23] um I did that. Tell me about Claude Co. [09:26] Yeah. Did you try that? [09:27] >> I have not. But uh before we jump into [09:29] that, I just wanted to say like uh I [09:31] have I've automated one thing now and [09:33] it's not 100% automated, but I made um [09:36] this thing. I was like, "Hey, uh I think [09:38] I used warp on this one, not open code [09:40] because open code had the thing where [09:41] you can't use your claw tokens anymore." [09:43] >> Yeah. [09:44] >> Um [laughter] [09:45] >> rugpool. Rugpull. Yeah. [09:46] >> Yeah. Um but I was using uh warp and I [09:49] was like, "Hey, um build me this thing. [09:51] I It can be swift because like it's just [09:53] going to be me that uses it. So it can [09:55] be Mac only. I don't care. Um, I want to [09:58] put in a link to a whiskey.fm episode [10:02] and I want you to pull the title, the [10:05] show notes, and the transcripts. Pull [10:07] like five to 10 topics. Then take those [10:11] topics and make them into like concrete [10:14] things you could make an image of. Not [10:16] just like React hooks like like [10:19] something that's like uh a React hook [10:21] doing like I don't know like make it [10:22] like a thing that could be drawn then [10:25] give it all back and like give me an [10:28] image that I can use as the art for the [10:29] episode cuz I'm trying to make like I [10:31] spend too much time on those for no [10:33] reason. So I'm like [10:34] >> they're super cool dude. It's not no [10:35] reason. I love those things. every [10:37] single one of them. I'm like, "Dude, the [10:38] TLC in this is awesome." [10:40] >> Yeah, but like I don't have the time for [10:42] that. So, they might get less good. [10:43] They'll they'll probably have less [10:45] people in them and be more topic based. [10:47] Um cuz I used to like try to describe [10:49] the person that was in the episode and [10:50] have it like draw what they would look [10:52] like or whatever. [10:54] >> Um but now I'm just going to be like, [10:55] "Hey, what do we talk about?" Like, [10:57] boom, we got it. Like better than having [10:59] no art, for sure. So, I'm trying that [11:01] out. Working okay so far. [11:03] >> Cool. I've used uh Warp to create an [11:05] automation for me on my personal blog so [11:07] that every time I drop an image into the [11:09] image folder, it sends it through my I [11:11] have like two phases of processing that [11:13] I do as well as generate the front [11:16] matter for me. So that was like a really [11:17] cool one to give to AI because the ins [11:19] and outs are so clear. Um [11:22] >> and then I saw something this last week [11:23] too that I was very impressed by which [11:25] is someone was um [11:28] making a whole website a a data driven [11:30] entire website. So, let's say you've got [11:33] uh let's just pick music. You want to do [11:35] a website for the top 20 songs of 2026, [11:38] whatever. [11:39] >> Okay? [11:39] >> So, you go get the data. You stick it in [11:41] uh you start a markdown file and you're [11:44] basically you build out this pretty [11:45] large prompt. It's almost like a big [11:47] PRD, but in the middle of the PRD, this [11:50] person stuck a API key to Korea and to [11:55] um Gemini Nano Banana and said, "Hey, I [11:59] I have no media for this site. Use these [12:03] keys and create images that are engaging [12:06] and um you know, help with the site or [12:10] whatever." And then they took this [12:11] humongous prompt which uh and yeah [12:13] generated images for the whole thing [12:15] which was anyway I thought that was cool [12:16] but they pitted it against a Ralph loop [12:18] um versus not a Ralph loop and the Ralph [12:22] loop um was arguably not as good [12:25] although it was really splitting. Some [12:26] people were like the Ralph loop one is [12:28] better and I was like I kind of like the [12:29] first one better. They were both really [12:31] good though. The PRD was just so well [12:33] written with all these little details [12:34] like I'm saying about like the image uh [12:36] generation and these API keys. I was [12:39] very impressed. Also very intimidated. I [12:42] was like, "Oh geez, this site looks [12:44] awesome and it got made for this person [12:46] within minutes." I'm like, "Crap." [12:49] >> Yeah. Yeah. I mean, I think sites [12:52] everything is doing better and better [12:53] at. I think images are still kind of hit [12:55] or miss. Like it especially if you tell [12:57] it not to do a thing because if you've [12:59] said the words, [laughter] if you're [13:01] like, "Do not put facial hair on this [13:03] person." is like, "Well, you said the [13:05] words facial hair, so I'm going to put [13:06] so much facial hair." Like, you can't [13:08] tell it not to. You have to describe [13:09] what you want and just omit the things [13:11] you don't want or it will freak out. Um, [13:15] but yeah, the next thing I want to build [13:16] is like [13:18] >> um we need a thumbnail generator. So, [13:20] there's one built in to Riverside, which [13:22] I've been trying um for like YouTube [13:24] thumbnails to be like like you [13:26] [laughter] know, like you need just the [13:28] random guy with the weird face like and [13:30] a title. Um, and so it's been doing [13:33] those and it won't change the text. I [13:35] even asked it. I was like, "Can you like [13:36] make the text not just the first word [13:39] highlighted in yellow and then the rest [13:41] black and like give me different fonts [13:44] and do different stuff?" It's like, "Oh, [13:45] absolutely. I have full control over [13:46] that." And I was like, "Cool. Can you [13:47] like change it to this?" And it was [13:48] like, "Cool. I did it." And it was still [13:50] the [laughter] same thing. I was like, [13:51] "Clearly someone has baked this into [13:53] your like base knowledge somewhere." [13:55] >> Hey, that's that AI harness stuff. We've [13:57] talked about that. It was like Grock. I [13:58] was, "Hey, Grock, I want you to make me [13:59] a meme.me. make me a spicy meme that [14:01] pisses off developers, you know, like uh [14:04] I know you know what makes people mad [14:05] and is is spicy. Do it. And it um it was [14:09] like hooks. It had some comment about [14:11] hooks and then um it couldn't even spell [14:14] react right uh in the in the image it [14:17] made. I'm like, "Ah, you're so stupid. [14:19] This is bad." [14:20] >> Yeah. Yeah. Anything like that there [14:23] isn't a ton of source material on it has [14:25] trouble with, too. Like what I did a I [14:28] had a tweet series for like I don't know [14:30] if I was trying to do one a day or one a [14:32] week or whatever it was. It was [14:33] unsustainable. I was trying to do like [14:35] dad jokes about programming and I was [14:37] like all right AI I've got like 10 [14:40] pretty good ones here like can you [14:41] understand like how I did these and like [14:44] try to use the same sort of patterns [14:46] pattern matcher. Yeah. [14:48] >> Yeah. And it was like Yeah. Yeah. Yeah. [14:49] And it gave me like a bunch more and [14:51] they were all awful. So, I was like, [14:53] "Oh, god." All right. There's just not [14:55] enough people making dad jokes about [14:57] programming, I feel like. So, [14:58] >> oh jeez. Um, I tried to give a layup to [15:01] So, this is Yeah, I like to describe [15:02] this as giving a layup to AI. You're [15:04] like, "Hey, AI, I'm spelling out [15:06] everything. You You're smart, right? Use [15:08] your intelligence to learn from what I [15:10] just taught you." And And it's like it [15:12] never works. [15:13] >> Um, so one of the things I have on the [15:15] agenda here to talk about is uh AIM. Um, [15:19] which is just a trendy acronym. It's not [15:21] even a What's your aim handle? Um [15:24] [laughter] [15:25] I'm trying to Well, I succeeded. I Do [15:27] you know what flip is? Um from that was [15:29] made by Paul Lewis. The acronym first [15:32] last invert play. Have you heard this? [15:34] >> I don't think so. [15:36] >> It was a way for you in JavaScript to [15:38] take an element that needs to it's like [15:40] in line in the page and now you're going [15:42] to make it fixed position like over [15:43] here. Uh, so if you're listening [15:46] everybody, I I'm holding my hands up in [15:48] two different places and one hand is [15:49] closed and one hand is open. All right. [15:51] So anyway, there's like an element needs [15:53] to go into two states. Uh, and they're [15:55] both [15:56] >> Tell me about sprites. Is one sprite [15:58] entering? [15:59] >> Yeah. [laughter] Oh, sprites. [16:01] >> No, I'm sorry to derail you. Keep going. [16:03] >> Oh, it's all good. Okay, so you got [16:05] element number one and you you like [16:07] click it and it needs to morph to the [16:09] position uh in position number two. And [16:12] it's not two different elements. it's [16:13] the same element needing to exist in two [16:15] different layouts and you want to [16:16] animate it between these things and you [16:17] want it to be a CSS transition so that [16:19] it's like interruptible and stuff like [16:20] that. Okay, so Paul Lewis comes up with [16:22] this uh awesome simple way to do it [16:24] called flip. And the F in flip is for [16:27] first you take the element where it is [16:29] in the closed state um at the beginning [16:31] and you take a it's not a picture but [16:33] you get its bounding um qualities. You [16:36] get it's like what's its top and its [16:37] left? What's its height and its width? [16:38] What are all these different properties [16:39] about? What's its background color? And [16:42] then you do the L, which is last. you [16:44] actually put it into the new layout [16:47] instantly [16:48] um by just setting the properties and [16:50] causing reflow. So then it goes into [16:52] that end position. You take your uh [16:53] measurements again and then you invert [16:55] it which means you put it back to the [16:57] first position and now that you know the [16:59] first and then that end position you can [17:01] interpolate between the two states and [17:02] you get this really nice morph effect [17:04] with JavaScript and a CSS transition. [17:07] David Corshe who we just had on a show [17:09] recently has done a ton of flip. He [17:10] wrote an article on CSS tree. Flip is a [17:12] very cool technique for very dynamic, [17:15] meaningful morph transitions. And I last [17:17] week figured out how to do it all in CSS [17:20] using anchor positioning, uh, [17:23] interpolate size, and oh crap, there's [17:26] one more feature. Oh, starting style. [17:29] >> And so I did this. [17:31] >> Yeah. Well, I called it AM stands for [17:33] anchor interpolated morph because the [17:36] okay, so if you think about it, it's [17:37] like a the first part was uh finding out [17:39] the measurements of the element, its [17:40] height and its width and its position. [17:42] Anchor can do that. You just set that [17:44] first element, you're like, "Hey, you're [17:45] an anchor uh element." And so now [17:47] anything else can like know the [17:49] positions and bounds of that of that [17:51] element. And then you can take the next [17:53] element which is just in its natural [17:55] resting place uh and then say starting [17:57] style start your styles from where this [17:59] ele other element exists. So I'm able to [18:02] make these elements morph from any [18:04] element in anywhere on the screen uh [18:06] using CSS because I can use anchor to [18:08] measure where they are. They know where [18:10] they are naturally and then they can [18:11] like morph from that position. Anyway, [18:14] it's really cool. So I'm like figuring [18:15] out how to do flip with CSS uh and [18:18] anchor. So Anchor is really cool and [18:19] anchor just went baseline in Firefox. [18:21] So, [18:22] >> I was just about to ask, is it supported [18:23] everywhere yet? [18:25] >> Uh, I mean, it's not like we can use it. [18:26] It's so funny. All these cool things [18:28] come out. I'm like, at work, I can't use [18:29] it. Even even when it's baseline at [18:30] work, I'm like, I it's I have to support [18:32] iOS. [18:32] >> Where is it? [18:33] >> Oh, cuz because of older stuff. So, I [18:35] don't [18:35] >> Yeah, for me it's like I support it [18:37] everywhere. I don't give a [ __ ] about [18:39] the older stuff. Like, if if the newest [18:40] browser that's not like a Canary [18:42] supports it, ship that. Like, [18:45] >> nice. Well, so you're Yeah, big fan of [18:46] baseline. My site kind of does that. [18:48] Well, my site will early adopt all sorts [18:49] of things. I'm willing to manage two [18:51] versions or even just have a lame [18:53] version and then a dope one. I'm like [18:55] that's that's cool with me. But at work [18:57] we uh we can't do that. It's like it [18:59] can't be broken. [19:00] >> Yeah. [19:01] >> Uh [19:01] >> yeah, work is different. I'm talking [19:02] about my personal stuff. I even did like [19:04] brley before it worked and like [19:06] everything like Chrome was the only [19:08] thing I think and I was like hey it's [19:09] way smaller going to ship it. Like [19:11] >> that's awesome. [laughter] [19:13] >> Yeah cuz like there's not that many [19:15] people coming to my personal site or [19:16] whatever. So, I'm like the people that [19:18] are are probably developers on like the [19:20] latest Chrome. Like, it doesn't doesn't [19:22] matter. [19:22] >> Um, [19:24] >> but yeah, I'm excited for the anchor [19:26] stuff because I've been wanting to use [19:28] that. Um, [19:30] >> like you can't do everything just yet. [19:32] You're going to need like some polyfills [19:33] or extra code or whatever to support [19:37] like we're using a floating UI um in [19:41] Shephard to do like basically what [19:43] anchor stuff does. Um, [19:44] >> yeah. And I would like to replace that [19:47] because it is big and it is like the [19:49] biggest part of the uh the library as of [19:52] right now. [19:53] >> It's also badass but yeah. [19:55] >> Oh it's great. Yeah. No shade on them. I [19:57] just like you know in the chase or not [19:59] chase race to the smallest libraries. Um [20:02] I'm trying to do that. And also I might [20:05] replace spelt with solid because okay [20:07] spelt five for whatever reason and Rich [20:09] is not helping me with this. So, Rich, [20:10] if you hear this, uh, maybe we can make [20:12] it better, but like it it increased the [20:15] bundle size by like 10 kilobytes or [20:17] something. Um, [20:19] >> like it went from like 17 to like 28 or [20:22] like I guess [20:23] >> four to five it went up. [20:24] >> Yeah. Mhm. And so I think what he was [20:28] saying is like the the wayfelt 5 works [20:30] is [20:32] like the base part like uh I don't go I [20:36] can't think of words. the like uh the [20:38] main part you're shipping like the [20:39] little what do you call that the core [20:41] the core the core is like bigger um but [20:44] then each component is smaller and since [20:46] we have like five components we don't [20:48] benefit from that because like [20:50] >> it's like if you had a like a thousand [20:53] component app then you'd be like oh my [20:55] god it's so much smaller because like [20:56] each component is smaller but we have [20:59] five components and so like it's not [21:02] smaller because the core is bigger which [21:04] I think is what we're hitting um we are [21:06] also still using the like legacy don't [21:09] use this classes thing because like [21:12] >> because it's uh it won't let you send a [21:16] reference to an element back out still [21:19] so I need some magic from someone who [21:21] knows how to do that to fix that um but [21:24] I was like yo AI um this is pretty well [21:28] tested you want to convert it to solid [21:30] for me and it was like dope and it did [21:32] and it was back to like the size of [21:33] spelt 4 and I was like hm could ship [21:37] like [laughter] [21:39] so um I'm interested but I I don't want [21:41] to like I like philosophically like [21:44] spelt so I don't want to like move away [21:46] from spelt because just because like I [21:48] want to make for sure that it's like [21:49] what I have to do before I do it. [21:52] >> Yeah, good call. You know the it kind of [21:54] goes into one of our other topics. It's [21:55] like uh you want to make the smallest [21:57] library. How about your library doesn't [22:00] even ship code? [22:02] >> Yeah. [laughter] See, did you see this [22:04] post I put in the Discord [22:07] as article somebody wrote? [22:09] >> No. Where they How would if it doesn't [22:11] ship code, what does it do? [22:13] >> Well, let's let's talk about that. They [22:15] ship the specs and um all the [22:19] expectations. [22:20] >> So, it's basically shipping instructions [22:23] for an LLM. So, here's the thing. this [22:25] person shipped a library that basically [22:26] does relative dates, you know, like 5 [22:29] seconds ago or 20 20 days ago and it's [22:31] all localized. It does all it does all [22:33] the things like a huge date functions [22:35] library would do. However, it ships no [22:38] code. It just ships the the great spec [22:42] and instructions and expectations so [22:45] that an LLM can do it in any language. [22:47] And so this person's like, I made a [22:49] library that works across Elixir and [22:52] Ruby and uh Rust and whatever because [22:56] it's not uh it didn't prescribe to a [22:59] language. It just is a spec. And I [23:02] thought this was super interesting [23:03] because I'm I'm very prospect driven [23:05] development. [23:06] >> Yeah. [23:06] >> And this was the first time I saw [23:07] somebody kind of do it at a micro level [23:09] where they're like, "Oh yeah, I'll just [23:11] ship all this stuff as a light." And I'm [23:12] like, "This is kind of awesome." Um, the [23:15] only thing that's like there are [23:17] downsides to it. Like you can't really [23:18] log bugs against the library, which also [23:21] maybe that's an upside. I'm like, "Oh, [23:22] no one can log bugs against me. That [23:24] [ __ ] rules." [23:25] >> Well, [laughter] you still could. I [23:26] mean, it's not a like, you know, uh, [23:29] can't read whatever of undefined bug, [23:31] but it's like, [23:32] >> yeah, [23:33] >> um, you know, the spec is maybe a little [23:34] wrong in this spot. Like, I want to [23:36] reward it like [23:37] >> add some tests for this case that the AI [23:40] keeps screwing up. Yeah. [23:42] >> Yeah. So, I think I think that's where [23:43] we're going because like [23:46] >> um people are already like not even [23:48] interviewing with code at all. They're [23:49] just like, "Hey, show me how you can [23:51] prompt and like let's see how productive [23:53] you can be. I don't care how you do it." [23:55] Um [23:56] >> Jake, stop. [23:57] >> Yep. Shopify is revamping and updating [23:59] their interview stuff. What's up, dude? [24:02] >> He's chewing up a box. He says, "I know [24:04] you're up here ignoring me." [laughter] [24:09] I can smell the dog breath through the [24:11] camera, bro. [24:12] >> Yeah. Yeah. Sorry. You were saying [24:16] >> uh something about [24:19] >> [snorts] [24:19] >> uh Shopify is updating their their [24:20] interview stuff, too. And I was asking [24:22] um and and I had and I have my own [24:24] answers to this, but like what if [24:26] someone shows up to the interview and [24:27] they [snorts] just uh Ralph Wiggum Loop [24:30] the whole question or what if they take [24:32] it uh give it to We were talking last [24:35] week about Gas Town. They just give it [24:37] to Gas Town. They're like, "Hey, mayor [24:38] of Gas Town. Here's a interview. Why [24:41] don't you just solve it?" And then they [24:42] just sit back and they twiddle their [24:43] thumbs. So, hey, what do you want to [24:45] talk about while uh all my agents do all [24:48] this interview? You know, like, you [24:49] know, do we reward that person? Do we [24:51] not reward the person? So, there's all [24:53] these different scenarios now that we're [24:55] It all depends. Yeah. So, basically, we [24:57] don't want anybody showing up that just [24:58] prompts the whole time and can't explain [25:00] anything that's going on. It's a bad [25:02] sign. [25:02] >> Or do you though? So, I mean, this this [25:05] is where I think we're going is like [25:06] there's going to be some percentage of [25:09] people. Hold on. Okay, hold on. I'm [25:11] going to shut Jake out. One sec. All [25:13] right, edit that out. Somebody editing [25:15] this later. Okay, [laughter] [25:18] so there's going to be some percentage [25:19] of people who are just like innately [25:22] uh sants at prompting, right? So, like [25:25] if they're for some reason just really [25:27] good at controlling all AI and they get [25:29] really good results that work really [25:30] well every time, do you care if that [25:32] person understands anything about the [25:34] code? If they're shipping insane amounts [25:37] of really good stuff that people who do [25:38] know about code go, "Wow, that's good [25:40] stuff." Like, I would say no. [25:43] [clears throat] [25:44] >> You're right. I hate [laughter] that. [25:46] But [25:48] >> it seems very fair in terms of like [25:50] they're producing the thing we want. The [25:52] output is desirable. Well, they can't [25:54] explain it, but who cares? They can't [25:55] explain why their products are good or [25:58] Yeah. [25:59] >> Like if you make good art, you're [26:01] probably like, "Hey, I don't I don't [26:02] know how I do it. I just go like and [26:05] it's good." Like [laughter] [26:08] >> some But then again, some pe some [26:09] artists who are like classically trained [26:11] are like, "Oh, I know all of the [26:12] different strokes and I like do it this [26:14] way and like you know, like [26:15] >> I think they're in their own colors, [26:16] right?" Yeah. [26:17] >> Yeah. There's room for both. And I think [26:19] it's not as much of a science as we were [26:22] led to believe. There's a lot of art to [26:24] coding and making an especially in web [26:27] development like in a uh you know purely [26:30] backend [26:32] runs on your like computer app or [26:34] something there's not as much art to [26:36] that because it's more just like input [26:38] output but like there's a lot of gray [26:40] area and stuff that has any kind of UI. [26:43] >> Yeah. I mean this is all very prospec [26:45] too because then the co because the co [26:47] we're just moving so quickly towards the [26:49] code matters less and less which also [26:51] comes in another one of the talking [26:52] points which I think we've briefly done [26:54] before which is uh does do types even [26:56] matter does having a database matter [26:59] because if a database is a layer of [27:00] indirection for all your data and AI [27:02] doesn't give a crap about that AI can [27:04] just build a site off of local files and [27:06] it can write to files and then you have [27:07] no dependencies like how much and where [27:10] is this going to end where code is is [27:13] completely irrelevant uh in favor of [27:16] >> yeah prompting and spec and spec [27:19] authoring. I even think the spec side is [27:20] even cooler. Like Toby [snorts] at [27:22] Shopify is has a lot of great um like [27:26] ideals and guidelines and just like [27:29] things that guide him in general. And so [27:31] you could kind of take a lot of that [27:33] stuff, codify it into like the heart of [27:35] the specs of your company and then have [27:37] all these additional people go build on [27:39] top of those and elaborate and then your [27:41] whole kind of system is a manifestation [27:45] of your ethos and your your services uh [27:48] kind of combined. And then you combine [27:50] that with brand and like a design system [27:51] and some other things and you're really [27:54] kind of reaching a new level of um just [27:58] producing software where the code [27:59] doesn't matter. it matters more about [28:01] the the core that you were talking about [28:02] earlier isn't the core of the framework. [28:04] It's like the core of like why why are [28:06] we doing stuff [28:07] >> uh becomes more important than what are [28:09] we making and that [28:10] >> I've I've always liked the why over [28:12] what. Um so I'm Anyway, this is Dude, [28:14] it's wild. [28:15] >> Can it slow down please? I'm I'm [28:17] exhausted. [28:18] >> Yeah, I wish it would slow down. Um, [28:20] this we were talking about this at work [28:22] is why I put this uh this bullet point [28:24] on here because we're just now in like a [28:28] a nine or 10year-old codebase adding [28:31] some TypeScript in which is always hard [28:33] in an old codebase with lots of Croft [28:35] and uh we were basically kind of having [28:38] a philosophical discussion around like [28:40] what do we want to recommend as the way [28:42] to do TypeScript right because you can [28:43] do it a few ways you can have like one [28:46] where it's like oh yeah I want [28:48] everything to be super strict Like if [28:50] you type a file, I don't want to ever [28:51] see an any. I want everything to be [28:53] perfect. I don't want to ever use a type [28:54] from that. That's wrong. There's another [28:56] philosophy that's like, well, I don't [28:58] care about most of this cuz it's pretty [29:00] trivial, but this one thing is really [29:01] hairy. I want to like type that and then [29:04] I want the rest to be any. I don't care. [29:06] Um, and we were just kind of talking our [29:07] way through like what do we want to do? [29:09] How do we approach converting the [29:10] codebase? And then we were kind of like, [29:11] wait, so you've got, you know, thousands [29:13] and thousands of files. It could take [29:15] years to convert. Maybe you can do a [29:18] code mod or some AI stuff to like make [29:20] it pretty quick. I don't know. We're, [29:21] you know, in the factf finding phase [29:24] right now, but we're like, but what is [29:26] the point anyway? Because if we're going [29:28] to like like the as an alternative to [29:31] converting anything at all to [29:32] TypeScript, what if you spent a lot of [29:33] time investing in your guardrails for AI [29:37] and you were like, "Hey, everyone on [29:39] like all these other teams who is maybe [29:41] not as senior or doesn't know the full [29:43] code base or whatever, like use this [29:45] when you're writing code and it will [29:48] keep you from doing dumb stuff that like [29:50] TypeScript would also have saved you [29:52] from, but like we don't care about [29:53] TypeScript anymore. We just want you to [29:55] like write everything through this. [29:56] It'll check and make sure like you're [29:58] not doing that. And then TypeScript [30:00] doesn't need to exist except for if you [30:03] fully typed it all too then AI could go [30:06] way faster because it could be sure [30:07] about being like oh I know all these [30:09] types are for sure this way. So like [30:11] there's still some benefits but it was [30:12] kind of like we're probably never going [30:14] to do that because why would we spend a [30:16] year of like a developer's time when AI [30:21] can just kind of save us? [30:22] >> Oh man, I'm in I'm in a code base all [30:24] day right now. Over a million lines of [30:26] TypeScript, no Annies. Uh, everything's [30:29] typed. [30:30] >> I love it. Jealous. [30:31] >> Um, [30:32] >> well, I I mean, there's just there's [30:35] just so many trade-offs. Like, okay, [30:37] first off, AI is not wildly empowered by [30:40] this. It's it's equally as hindered as [30:43] like a as like a human is. And also, I [30:45] think it might actually be adding [30:46] overhead. [30:47] >> Um, because it it churns. It doesn't [30:52] know. All it knows is like, I'm gonna [30:53] try to go do this thing. And it goes and [30:55] does the thing and it's like, oh, [30:56] there's a lint error. Oh, there's a type [30:57] error. And then it just like bounces [30:58] back and forth in there. And then it [30:59] still can't see. And even if you teach [31:01] it to see, um, it doesn't Oh, man. We [31:05] We're trying to turn this thing into [31:06] like a little human. Um, and I'm not [31:10] seeing it wildly do great things inside [31:13] of a big codebase. And we just Okay, so [31:15] here's a good example. We just saw [31:16] [snorts] uh codeex 5.2 come out. Uh, [31:19] chatgbt codeex, right? Yeah. That's open [31:22] AI codeex anyway. And its whole thing is [31:24] they're trying to pitch, hey, if you got [31:26] a big ass codebase and you finding that [31:28] AI models are really shitty at going [31:29] through it and having longunning ideas, [31:31] we just made the AI model for you. It's [31:34] called Codeex 5.2 and it's specialized [31:36] for longunning [ __ ] And it's like [31:39] [laughter] [31:40] >> I don't think it's actually going to do [31:41] that much better, man. Like it's kind of [31:43] like there's so much code and there's so [31:46] many edge cases and it's like CSS. CSS [31:48] constantly shows up and it's all like [31:50] it's all like my job's done. You're [31:51] like, "Did you look at it?" And it's [31:52] like, "Yeah, I took a picture. Look at [31:53] the picture. Isn't it pretty?" And [31:54] you're like, "It, no, it's vomit, dude. [31:56] Like, did you even know?" And it's like, [31:58] >> and then the other thing, too, is like [31:59] you're trying to do code mods or [32:01] migrations or whatever it is. Like, it [32:02] can take a picture of the beginning [32:04] state and it'll take a picture of the [32:05] end state happily and be like, "I took [32:07] pictures." And you're like, "Yeah, but [32:10] what [32:11] >> do they look the same?" [laughter] [32:13] >> Yeah. Well, a lot of times they're not [32:15] supposed to. So, it's like how do you [32:16] how do you describe what's supposed to [32:19] change in the picture? That's the [32:21] follow-up one. If it's a layout change [32:25] based on like the uh [32:27] >> Oh god, my brain doesn't work. Like the [32:30] visual diffusion or whatever, wherever [32:32] it like, you know, makes it all noise [32:34] and then the picture. [32:36] >> Yeah. So, like if we can figure out how [32:38] to tell that like this part of the noise [32:40] is now blue, it could be like, "Ah, yes, [32:43] I've I've made the noise a noisy blue [32:46] right there and it is now blue." Um, [32:48] [laughter] I don't know. I feel like [32:49] there's a way to do it. [32:51] >> We got comments in here from Mr. Ray [32:53] Dev. He says, "Codeex is crazy, worse [32:55] than even CL." So, Mr. Ray Dev was like [32:57] a long running, let me try it. And then [32:59] all of a sudden, it's not. And I'm just [33:02] not surprised by that cuz um I I want [33:05] them to be good. And I think I talked [33:06] about it last week too where I've been [33:08] rolling forward a skill. So that way [33:09] when I spin up agents that um you know [33:12] like I have one little agent does this [33:14] task and it helps evolve a skill and [33:16] that way it passes information forward [33:17] and I don't have to keep a database. I [33:18] don't have to keep beads. I don't have [33:20] to keep git. I don't have to have gas [33:21] town. It's like I just have this one [33:23] migration. I don't need one code mod. [33:25] I'm not trying to do this to my whole [33:27] and it still um it just screws up enough [33:32] times and it also hiccups on types [33:34] constantly. So that's why I'm looking at [33:35] TypeScript going, what are you really [33:38] doing right now other than perceptively [33:41] slowing us down? Um, [33:43] >> yeah. Yeah. I mean, I'm I think using it [33:46] as a llinter and not trying to be super [33:48] strongly typed and using it like [33:50] sparingly, I think is the way I would [33:53] recommend using it now. Like if it's a [33:55] project that just I'm working on, I'm [33:57] going to I'm a completionist with the [33:58] type, so I'm going to get it all typed. [34:00] But like I think in terms of you know [34:02] leading a team just type the like one or [34:04] two things that you know people are [34:06] getting wrong a lot or whatever like [34:08] help them there but other it's just a [34:10] llinter like you could also write a [34:12] eslint rule that would like check am I [34:13] using this thing wrong and like you know [34:17] I feel like it's like that it just has [34:18] less steps. So like I think that is the [34:20] way I'm going to probably be using it [34:22] going forward versus trying to have a [34:23] fully typed everything. Um, although I [34:26] love when things are fully typed, but [34:28] >> fully typed feels good. I'm not gonna [34:30] lie, dude. It always That's why Rust is [34:31] popular. If you can make your Rust app [34:33] run, you're like, "Hell yeah." [34:35] [laughter] [34:36] >> Per like a kitten. Uh, you know, like [34:39] there's no mess ups there. I feel like [34:41] AI though, like h just it once you're [34:43] past like four files, [34:45] it just starts to go downhill. And even [34:47] when they run these things overnights [34:49] and weekends, like people are just [34:50] spending tokens. And I'm not seeing [34:53] amazing results yet. And so here's the [34:55] thing though is like I'm assuming it's [34:57] going to get there. And so um I'm going [34:59] to keep tinkering. I'm going to keep [35:01] trying to see how these tools work. Um [35:04] because they're not done growing every [35:05] week. It's just wild. So we'll see where [35:08] it all goes. [35:09] >> Yeah, it's been pretty good for me. I've [35:11] been So at work we have an exception to [35:14] use co-pilot and only co-pilot [35:16] basically. Um [35:18] >> sucks to be you. [35:19] >> Well, however, so we have a loophole [35:22] now. This is new development as of [35:24] today. Um, GitHub tweeted like open code [35:27] is Jared tweeted [35:30] >> fully supported and I was like, hm. So I [35:33] was like, okay, is this a loophole where [35:35] we can use open code now? And the [35:37] principal on my team was like, [35:39] >> yes. And I was like, hell yeah. All [35:40] [laughter] right, we're [35:42] >> you got to go install that. Oh my open [35:44] code. It's uh a really great layer on [35:46] top of open code. I'm a big fan. [35:48] >> So I want to come back to that and hear [35:49] what's in that. But did you see they [35:51] have a app now? Like it's not just the [35:53] CLI. You can install an open code app. [35:56] >> I saw you write that and open code [35:58] black. So inform me. I am not caught up. [36:00] >> So open code black is this thing. Um [36:03] it's like uh the claude code like or cla [36:07] whatever where they have like the $20 [36:08] tier and then the max tier that's like [36:10] $100. It's like their version of that, [36:12] but it just like spreads the work across [36:14] whatever model they want at the time. [36:16] So, it's like Open Code Zen kind of [36:18] except with a like more bakedin price [36:20] instead of charging you per token. Um, [36:24] and so people were signing up for it cuz [36:26] they would open up like 20 spots at a [36:28] time and they would sell out like before [36:29] anyone could click it. [36:31] >> So now they have just a wait list like [36:32] you can go buy it and be like we'll [36:34] charge you whenever we like onboard you [36:36] finally. Um, [36:38] >> so I signed up for for it and I I'm [36:41] gonna give it a try. Um cuz open code I [36:44] think is the best uh developer [36:46] experience thus far. So I want to try it [36:49] with a like I don't have to worry about [36:51] every couple hours it going hey you got [36:53] to wait a while to like do anything [36:54] else. Um and also [clears throat] like [36:56] we said you can't use it with I was [36:58] using it with my anthropic like $20 code [37:01] which no longer works. So you can use [37:03] anthropic like charge me based on every [37:05] character you send me. Um but I don't [37:08] love that. So [37:10] >> yeah, dude. Let's uh let's compare. [37:13] Let's wait. Open code black cornite for [37:16] dead. [37:17] >> Nice. [laughter] It's good. [37:21] >> I'm trying to like So wait, what is it [37:23] that you like about open code over [37:25] claude over [37:28] um uh warp for example? Or even like are [37:31] you in cursor or are you in windfer [37:34] and windsurf are dead to me. I don't [37:36] want those. [37:36] >> Okay. So you're terminal all day. No [37:38] idea. [37:39] >> Are you using work trees? [37:41] >> No. [37:42] >> I'm not doing anything super advanced. [37:45] >> Um, and a lot of my preferences are just [37:47] vibes. Like was the first one [37:51] >> was to do the LSP stuff. So it like it [37:54] was better at TypeScript than everyone [37:56] else just cuz they connected it to a [37:57] TypeScript LSP. [37:58] >> And so that won me over. Like that was [38:01] the first thing where I like told it to [38:02] do something and it gave me a thing that [38:04] I would actually ship. Like I did a lot [38:07] of AI stuff before and it was like ah [38:08] this was like an interesting you know [38:10] debugging exercise but I'm not going to [38:12] ship your code and this one I was like [38:14] wow you did a good job. So I I like [38:16] liked open code for that. Warp has done [38:20] a good job as well but it seems to like [38:24] I don't know what it is. Maybe it's [38:25] because it's kind of like I have it set [38:27] to autopick the model or whatever. Maybe [38:29] it's picking bad models. Um but it's [38:32] like very hit or miss. Like sometimes it [38:34] does a terrible job. Sometimes it does a [38:35] great job and it feels kind of like AI [38:37] used to feel like um so yeah sometimes [38:42] I've I've been using warp more [38:43] especially since the claw stuff stopped [38:45] working in open code um and I've gotten [38:47] some good stuff out of warp. Warp [38:48] actually fixed a bug for me today [38:51] >> with uh in swatch there's a [38:54] >> um magnifier color picker thing. I don't [38:56] know if I talked to you about this bug [38:57] before, but so you got this circle, [38:59] right? And you move it and the middle [39:01] pixel like picks that color. And on [39:03] Windows, it works if you have a just a [39:07] 100% DPI screen, like there's no scaling [39:09] factor. But if you I have [39:12] >> 72 or or what do you mean? [39:15] >> Like like your your screen resolution. [39:18] Let's say you have a monitor that's like [39:20] 10. [39:20] >> Oh, it's at it's at its natural natural [39:22] resol. Yeah. Yeah. So then it worked. [39:25] But if you like I have mine set to 200% [39:28] because I have like a 5,000 pixel [39:30] monitor and so if you don't it's like [39:32] teenytiny and I can't read anything. Um, [39:35] so at 200% [39:37] it was not it was like a thousand pixels [39:39] off like it was clear offset like wasn't [39:43] factoring in the DPI and uh so I had [39:46] asked Open Code a bunch of times to fix [39:48] this like I'd released several versions [39:49] and it kept being like it's fixed ship [39:51] it and I shipped it and it didn't change [39:52] anything and um so Warp was like went [39:56] down the same path for a minute and was [39:57] like oh yeah let me fix it like we got [39:59] to see what we multiply and divide and [40:01] whatever and then was like wait what if [40:02] we just make the windows side DPI aware. [40:06] So we don't have to do any math. We just [40:07] let it like pick the pixels. I was like, [40:09] >> why haven't why has no one said that [40:11] before? Why is that's an option? [40:13] >> Um so so we did that and it works now. [40:16] So like warp saved me there. I don't [40:18] know what model it was using or but like [40:20] yeah it's it's like I feel like warp is [40:23] right 50% of the time but does like [40:25] super magical good stuff when it's right [40:28] and open code is like uh stuff you could [40:32] ship more often but is maybe not as [40:34] magical and fancy and good. I don't know [40:36] like it's it's more the baseline that I [40:38] can kind of count on. Um I think there's [40:40] room for both and I've I've had success [40:42] with both. [40:43] >> Yeah. So open code I've been playing [40:46] with that for a couple weeks. Uh it's [40:48] now available at Shopify uh as well as [40:51] cloud code. Both of which I can kind of [40:52] pick whatever model I want. And so I [40:55] have a lot of experience with both of [40:56] those. Oh yeah, we'll talk about oh my [40:58] open code here in a sec. But I also use [41:00] cursor a lot. Um cursor has some killer [41:03] features. Like people are like uh I I [41:06] used claude today to to spin up three [41:09] agents to do some work. And I'm like you [41:10] know in cursor that's just a dropown [41:12] menu, right? You you write your prompt [41:13] and you say how many agents do you want [41:15] to do the work? You say three [41:18] >> and it creates work trees, bro. So, it's [41:20] even fancier than that. So, you go like [41:21] three. [41:22] >> I haven't used it in a while. When I [41:23] used it, it was like open-ended text [41:25] box. I was like, "Can you make this [41:27] change?" And it was like, "I made it." [41:28] And I was like, "No, you didn't." It's [41:30] like, "Yeah, I did." I was like, "No, [41:31] [laughter] you didn't." So, I I didn't [41:34] Maybe I should try it again because it [41:35] was terrible when I used it last. [41:36] >> It's also like pretty wildly different. [41:39] Well, it has two modes. Same thing with [41:40] anti-gravity. Like, they have a agent [41:43] modes where you're like, I don't even [41:44] look at the code. you're almost purely [41:46] in like a but I don't use those as much [41:48] but anyway yeah if you if you open up [41:50] cursor fly out the side menu write your [41:52] prompt look for the one X that's [41:54] underneath your prompt change it to [41:55] three or four and all that does is [41:58] create agents and work trees for each of [42:01] those work streams they work you get to [42:03] watch their little batons come back and [42:05] then they highlight green you click them [42:07] and you see their diff [42:09] and it changes your code as if it would [42:11] have done before and then you can like [42:13] either accept it or try on the other one [42:15] and try on the other one. So every time [42:17] you click one, it changes your code to [42:19] their changes. You can see it in your [42:20] browser if you've got live reload going [42:22] on. Figure out the one you want, accept [42:24] the one, delete the work trees, and go [42:26] back to now a bunch of possibilities and [42:30] pick the best one. [42:32] >> Yeah. I mean, okay, wait, and I rephrase [42:33] like cloud and open code can do that [42:35] exact same thing. It's just you have to [42:37] be very explicit and deliberate and you [42:39] don't get the really rich switching kind [42:41] of experience [42:42] >> cuz like open code definitely uses sub [42:44] agents cuz I've seen it say like I've [42:46] got a bunch of sub aents doing this [42:48] stuff but it doesn't let you see [42:49] >> and pick like [42:51] >> it's not doing them side by I think it's [42:53] doing like separate things. It's like [42:54] four agents doing four different things [42:56] not four of the same to see which could [42:58] have the better output. You know what? [43:00] So I think that's a compelling [43:02] >> use case because you do get wildly [43:03] different stuff. It's nondeterministic. [43:05] So, like if I had four things doing the [43:07] same thing, I picked the best one. I [43:08] like that. I might have to try that. [43:10] >> Yep. There's also um there's so many [43:13] this this world is crazy right now. [43:14] Okay. So, the other thing is you have [43:15] people uh and [snorts] I've been trying [43:16] this too where you pit different AI [43:19] models against each other. So, no longer [43:21] like right now we're in a very mono [43:23] model. Oh [ __ ] that was M and M and M. [43:27] Uh right. You're like monogamous with [43:29] your model. But a lot of there's people [43:31] out there that are polyamorous with [43:32] their models. That's just fun to say. [43:36] Um, and they uh are basically saying, [43:38] "Okay, hey Gemini, you go do this task. [43:41] Hey Claude, you go do this task. Hey, [43:43] Chad, you go do this task." And then, [43:44] "Hey, Claude, go review this work that's [43:48] that came back from Claude." Oh, or [43:50] whatever. Anyway, you have them review [43:52] each other. Um, and that's like one of [43:53] the superpowers of open code is that you [43:55] can kind of pit them and warp, right? [43:57] You can open up you can open up command [43:59] T and warp. You hit command D to split [44:00] it vertically. Command shift D and now [44:02] you got four quadrants like you're in [44:04] T-m. You can have four different agents [44:06] reviewing other agents work to sort of [44:08] get yourself into this scenario where [44:10] you're not trusting a single agent to do [44:12] everything. That's wild. And so yeah, [44:15] you combine that with like work trees [44:17] and these other things. So here's the [44:18] magic of warp that I still really like [44:20] is that um I don't have to care about [44:22] what is an MCP server, what is a skill, [44:24] and what is a hook and what is a [44:26] command. I get to tell it what I want [44:29] and it sort of deres the best thing for [44:31] me. And my favorite thing is still MCP [44:34] discovery where I'm like, "Hey, I need [44:36] this thing to to get done." Uh, like [44:39] doing that, too. [44:40] >> It was like I was like, "Do some git [44:42] stuff." And it was like [44:43] >> pulling the git kraken mcp to figure out [44:46] how git works and I was like, "Oh shit." [44:48] Okay. [laughter] [44:51] >> Yeah, that's really cool stuff cuz uh [44:53] it's really annoying configuring things [44:54] like it's okay. Let me let me rephrase. [44:56] Configuring things is very fun. It's [44:58] also a time waste. So like if something [45:00] can kind of configure on the fly based [45:02] on my request, that's really cool. Uh [45:04] and Warp is the one that constantly does [45:06] that for me where I'm like I I ask for a [45:08] thing and I say, "Hey, go verify it." [45:09] And then it'll be like, "Oh, sure. I'll [45:11] spin up a browser and all of a sudden [45:12] Chrome opens up with this special [45:14] profile and I can see it taking pictures [45:16] and then it starts to cost me a lot of [45:17] money because it's like, I took pictures [45:19] and then I took the picture and I [45:20] tokenized it and I sent it to the server [45:22] and I reviewed it and I'm doing it again [45:23] and again and again." I'm like, "Stop [45:25] taking pictures right now. You're going [45:26] to eat up all my credits." [laughter] [45:27] >> $20 later. You should use this MCP. [45:30] >> Crap. Yeah. [45:33] But yeah, that's uh I was trying to just [45:35] like tell someone why I like Wind Surf [45:37] better than Cursor the other day and I [45:40] just couldn't. It was vibes and I'm like [45:43] it's annoying. I wish I wish I could [45:44] tell you. [45:45] >> I had both and that was the same for me [45:47] too. It was like um at the base use case [45:50] of like you know opening a file and just [45:54] working on some stuff and hitting tab to [45:56] tab through stuff. It was better. I [45:59] don't know why they both do the same [46:00] thing like but it was just I felt better [46:03] with the like tab autocomplete and at [46:06] the time the agentic stuff in both was [46:09] abysmal so it didn't matter. But like [46:11] the you know dayto-day I'm still writing [46:13] code. I just want you to autocomplete a [46:14] couple things worked way better [46:16] >> in Windsurf for me for whatever reason. [46:19] >> Okay. So let's talk about co-work and [46:20] then I want to talk about agency. What [46:22] do you know about I saw co-workers like [46:25] >> making files for people and has file [46:26] system access and I was like nope I am [46:29] not ready for this thing to like just be [46:31] reading all of my [ __ ] No thank you. I [46:33] want to give it to it with a spoon. I'm [46:35] not ready for automated. [46:36] I haven't used it, but I just saw [46:38] somebody tweeted like, "Thank God [46:41] anthropic is coming for everyone else's [46:42] jobs now." [laughter] [46:44] And I was like, "Yes, yes, that is uh" [46:46] cuz like I do find myself being like, [46:49] "All right, I have a, you know, I can't [46:50] think of an example, but some task I [46:52] want to do in life. Like, I'd like to uh [46:55] plan this thing, schedule this thing, [46:57] uh, you know, research getting tickets [47:00] to something, doing whatever." Like I [47:02] would like to be able to do approach [47:03] that in the same way of just being like [47:05] hello agentic thing like I want to do [47:07] this stuff tell me the best way like [47:09] find it for me. So I think it's [47:12] compelling to like apply that workflow [47:14] to the rest of your life but like yeah I [47:18] I do think you're going to hit a token [47:20] problem and like if we're using it for [47:22] everything uh the cost is going to catch [47:24] up at some point like you can't get used [47:27] to using it for everything. Um so I [47:29] don't know. I'm cautiously optimistic. I [47:30] think it's cool. I want to try it out. [47:32] But I just I just wanted to mention that [47:33] it exists. I don't have much else to say [47:35] about it as of now. [47:36] >> Me neither. I kind of got scared by it. [47:38] I was like, I'm not ready for that. You [47:39] know, the thing that scared me the most [47:41] though wasn't [47:42] co-work kind of like taking those jobs. [47:44] It's that I saw uh AI is now growing [47:46] crops. It's growing food. Uh and I was [47:50] like, "Oh, so farmers, maybe we won't [47:53] have farmers anymore." I was like, "What [47:56] if AI disrupts the farming industry?" [47:59] Yikes. [48:00] >> Yeah. Well, the farming industry is a [48:02] whole another episode we could talk [48:03] about. Like I think that whole thing is [48:05] broken. We should not have big giant [48:07] farms where like if one company has like [48:10] a bad growing season, we don't have [48:11] food. Like the local farm model is what [48:14] we should be doing, but I'm not I'm [48:15] going to get off my soap box cuz we're [48:16] going to run out of time. So, um yeah. [48:19] What? Tell me about agency or whatever [48:21] you were saying. [48:22] >> Okay. Okay. So agency um in in an age uh [48:25] where [48:27] u uh intellectual stuff is on tap uh [48:31] what you need now are people that are [48:33] adventurers and tinkerers and the people [48:36] that you can trust to go figure it out. [48:39] Um, so you need to we're seeing a rise [48:43] of agency, meaning that we want to give [48:45] people less of a specific role and trust [48:48] people to just go around and do good [48:51] like work for the company. You can focus [48:54] on this area. Uh, and we're not going to [48:57] tell you very much what to do. What we [48:58] want to do is trust you that you're [49:00] equipped with AI and that you are just [49:04] constantly doing good. You're you're [49:06] deploying agents over here to do that. [49:08] that's improving whatever you're over [49:10] here improving internal processes. [49:11] You're making an internal tool like you [49:13] are free now. You're you're now [49:15] liberated. Your chains have been broken [49:17] from your front end role or even your [49:20] design role. We want to give you agency [49:23] and and trust you to to just go do good. [49:27] And I'm like that's really cool. [49:29] >> So it's cool. But like in that world, [49:33] and I guess this is the world we're in [49:34] everywhere, even if you are specialized [49:36] in one specific area, like how do you [49:39] ever get a raise or a new position or [49:42] like if you're just kind of like you're [49:44] in charge of the vibes of this area of [49:46] work? It's like, well, how do I show [49:49] that I did good at the vibes? Like [49:50] >> you used your agency and you actually [49:54] changed. It's like [49:55] >> Yeah. [laughter] [49:56] >> Okay. Well, yeah. So, I mean, I guess [49:58] there's that. Like if you're if you're [50:00] your own manager in your own department [50:03] and you're setting your own goals and [50:05] you're writing RFC's that you comment on [50:06] yourself and you're like [laughter] [50:09] I don't know like I feel like there has [50:12] to be someone at some level that's [50:14] deciding like is the criteria you came [50:16] up with good? Do we agree? Did you do [50:18] well? Should you get more money like or [50:22] I don't know. Um maybe all of that is [50:24] done. Maybe there's no more promotions [50:26] for anyone ever. I mean, I feel like [50:28] we're pretty quickly heading towards uh [50:31] we don't need developers, designers, or [50:34] any of us anymore. So, that's [50:37] depressing. [50:38] >> Yeah. Please give me a severance of the [50:41] rest of the money that I would have made [50:42] for the rest of my life, and I'll be [50:44] happy. [laughter] [50:48] Oh, yeah. My my 401k is not not ready to [50:51] have no job. [50:52] >> [laughter] [50:53] >> It's going to be wild if uh if it really [50:56] does displace as many things as it's [50:57] kind of positioned to. [50:59] >> Yeah. No, I'm I'm trying to get in the [51:01] physical goods game. Like I was talking [51:03] to my friend who is a local farmer. He [51:05] has popcorn on the cob that he just like [51:09] he doesn't take it off the cob. You just [51:11] get the whole cob. He gives you a bag [51:13] and you put it in the microwave. I [51:14] thought you had to take it off the cob [51:15] to pop it, but you don't. [51:17] >> And so I was like this is so cool. Like [51:20] people would love to take a [ __ ] Cobb [51:22] and put it in their microwave. Yeah. [51:25] >> Why does no one sell it in this form? So [51:27] I'm like, "We should do that." Like why [51:30] are we do that online? [51:32] >> Yeah. Um and then he starts scheming. [51:34] He's like, "Oh yeah, you just need 40 [51:36] acres and you can grow 1 million cobs a [51:38] year." And like [laughter] [51:40] I was like, "Okay." Like I'm I'm not [51:42] ready for that. But like yeah, I want to [51:44] get into something like that. Like [51:46] experiment. Like I think AI can help you [51:48] iterate on your ideas and you're like [51:50] have a good website with basically no [51:52] effort and like throw up some Instagram [51:55] ads, see where it where it lands. Like I [51:58] don't know. I I want to get into more of [52:00] that kind of stuff because I feel like, [52:02] you know, the developer stuff can't [52:04] can't last forever. Like I think we'll [52:06] outpace most hopefully like before we're [52:09] stay. But [52:12] >> yeah, I don't know. It's bleak. It is [52:16] oddly bleak. Um I mean last week was [52:18] bleak, right? Now we're bleak still. I [52:20] don't know where [52:21] >> we always will be. We We still bring the [52:23] joy though. We're having whiskey. It's [52:24] okay. [laughter] [52:26] >> Hey, I showed a zombie killer. Oh, wait. [52:27] I want to talk about something that [52:28] brought me joy this week. It's uh want [52:30] my MTV.verell.app. [52:33] >> What? [52:35] >> It's you can watch MTV. So, do you [52:37] remember um [52:38] >> MTV just stopped running? Right. I saw [52:40] that. Like [52:41] >> Well, that's what I'm saying. It It [52:43] doesn't have to stop. You can watch [52:44] someone's curated everything off of [52:46] YouTube, all the music videos, put them [52:48] organized by like era and commercials [52:51] and put it into a thing. So it's like, [52:52] did you ever watch Nickreboot.tv? I [52:54] think it was Nick Reboot.tv. [52:56] >> You could just go to this website and it [52:58] was playing stuff and you just tuned in [53:01] like it was old school TV. There was [53:03] commercials, you know, skip it, skip it [53:05] or whatever. Um, [laughter] and you'd [53:07] watch Hey Arnold and you'd catch half [53:09] the episode because you tuned in late, [53:11] you know, and it was awesome. [53:12] >> Yeah. uh got shut down for probably [53:15] reasons. But [53:16] >> uh this MTV one is awesome in the same [53:18] way. You can be like, I want to listen [53:19] to 2000's um music videos and you just [53:22] go there, you hit a button, and there's [53:23] like 3,000 [53:25] >> music videos from the 2000s that'll just [53:26] randomly play for you and play [53:28] commercials in between every It's been [53:31] >> awesome, dude. And I'm finding new [53:33] music. I'm like, how did I not know that [53:34] song? That metal song was badass. [53:36] >> I was always a Fuse listener. Did you [53:38] ever listen watch Fuse? I didn't, but my [53:41] spouse was huge in a fuse. Yeah. [53:42] >> Oh, yeah. Yeah. It was uh like as soon [53:45] as that came out, I don't know what it [53:46] was like the alternative MTV, like more [53:48] rock based. And I just had it playing [53:50] like all day. Like I thought it was so [53:53] cool. [53:53] >> Um but yeah, I'm I mean music videos I [53:56] feel like are a lost art. Like people [53:58] are still making them, but it's like I'm [54:01] I personally don't like to go consume [54:03] YouTube. Um I'm probably in the minority [54:05] there, but like I like a thing that just [54:08] picks it for me, curate it. I don't I [54:10] don't want to go like click on it. I [54:11] just want it to be a a feed with [54:13] commercials and like Yeah, like [54:14] traditional TV sounds good. [54:16] >> You're going to love this. It's been [54:17] really nice for while I'm working. So, I [54:19] just put it on. Although, it's been it's [54:21] okay. So, let me rephrase. It's been [54:23] really nice. That's still the phrase, [54:25] but also sometimes really distracting [54:27] cuz I'm like, "That music video is [54:28] really sexy." I'm like, "Holy [ __ ] that [54:31] was that I watched that as a [54:32] 14-year-old. I'm like, goodness." I'm [54:35] like, "That [ __ ] was hot as hell." Jeez [54:37] Louise. [54:38] >> [laughter] [54:40] >> So anyway, I just from like watching [54:41] these music videos like that was okay. [54:44] Holy jeez. Wow. Um but also some of them [54:47] are just hilariously like simple and [54:50] awesome. So [54:51] >> yeah, [54:52] >> enjoy everybody. We'll put that link in [54:53] the show notes. [54:54] >> Yeah, we always end up uh watching music [54:57] videos at the end of like anytime we [54:59] have people over and we're hanging out [55:00] and it's like 1:00 a.m. That's like just [55:02] throw on music videos. So that'd be [55:03] cool. Just throw that on. You don't have [55:04] to even pick the next one. Yeah. Yeah. [55:06] Good call. So yeah, I uh got sixish [55:11] minutes left. Um I do want to mention [55:13] the topic that we didn't get to. Did we [55:15] get to it? And then I was like, should [55:17] we talk about it more next time? I don't [55:18] know. Anyway, the fact that booze and [55:20] nicotine and like, you know, any sort of [55:23] like thing that are fun or historically [55:26] thought of as fun um are like [laughter] [55:28] way down. Um [55:30] >> yeah. [55:30] >> And it's like people probably still like [55:33] fun. So, I don't know if it's like a one [55:35] one thing could be like a health thing. [55:37] Like maybe people I know there's a lot [55:39] of different like I get so many ads for [55:41] like this cocktail alternative it makes [55:43] you feel good. I'm like feel good how [55:45] what drugs are in it? Nothing. Okay. It [55:47] probably doesn't do anything. Like [55:49] [laughter] [55:49] >> so I'm like um I don't know. I'm [55:52] skeptical about those. I don't know if [55:53] you've tried any of those. [55:53] >> The NA party beer. Oh, I keep getting [55:55] Yeah, I keep getting those like, "Oh, [55:57] you still want to party but don't want [55:58] to drink? Here, have this NA beer." [56:00] >> Beer. Beer is awful. I don't want to [56:01] drink beer if it doesn't have alcohol in [56:03] it. That's stupid. Um, [laughter] [56:05] so no, but like the like kin euphorics, [56:07] have you seen those? Like no. [56:09] >> Um, [56:10] >> it's it's just like a herbbally I don't [56:12] know, neutropic, like supposed to be [56:14] like good for your mind, make you more [56:16] social, etc. Like you know, the good [56:18] parts of drinking, but like not the bad, [56:21] but like I haven't tried them, so I [56:22] guess I can't [ __ ] on it until I try it. [56:24] But I am very skeptical. Um, but anyway, [56:27] people could be doing some of that, [56:29] could be doing whatever. [56:30] How could that succeed though? If if [56:32] actual booze is down, how could the NA [56:35] alternatives be up? [56:38] >> I don't know. But yeah, sorry. Anyway, [56:40] keep going. [56:40] >> My hypothesis is that everything is [56:42] down. I think [56:44] >> everyone is losing their jobs. The [56:46] entire economy is down. Inflation is up. [56:48] Recession big time. [56:50] >> Minnesota's throwing bombs into people's [56:51] cars as they're trying to exit ICE. Yep. [56:54] It's nuts. [56:55] >> Yeah. Like um I saw on the Today show [56:57] the other day that pizza is down. like [57:00] 30%. [laughter] [57:02] >> Like what's [57:04] >> Yeah. Yeah. [57:04] >> That's what I'm saying. It's like if [57:06] even pizza can't do well, like pretty [57:08] much an American staple. Like [57:11] >> um I think everything is down because [57:12] everyone is like I can't afford to live [57:14] much less go get a [ __ ] pizza. So [57:17] like I'm going to make a pizza at home [57:18] if I want a pizza. Like I think that's [57:20] what's happening. So I don't think [57:22] people like I think there could be [57:24] increased demand for like all this [57:26] stuff. Again, I'm not sure. I'm just [57:28] like, you know, spouting whatever. But I [57:31] that that's how I feel. I I don't think [57:33] that it's like doom and gloom around the [57:35] alcohol industry and it's going to fully [57:36] go away because like you got to replace [57:39] it with something. People need something [57:40] to numb their minds. Um [57:42] >> tea, man. Just a just, you know, an [57:44] herbal tea. Um I keep wondering about [57:46] supply and demand. So as soon as I'm [57:48] like, "Oh, these things are down. So [57:50] supply is there, demand is down." I'm [57:53] like, "What? So is everything going to [57:55] get cheaper?" That would kind of [57:56] >> Well, that would be a welcome thing. [57:58] Yeah, cuz the whiskey industry has [58:00] gotten nuts. This stuff is way too [58:02] expensive. Like Chuck was telling me [58:04] about it and uh I don't know the full [58:06] story, but the the essential facts are [58:09] basically like uh you know they sell [58:11] like this whistle pig, let's say they'd [58:12] sell it for 20 bucks, 30 bucks, [58:14] whatever, like the original cheap price. [58:17] And then they were noticing people would [58:18] buy all of them and resell them on like, [58:21] you know, I bought thisl liquor.com for [58:24] like 60 bucks. And they're like, wait, [58:26] you guys are buying all this cuz you [58:28] think it's good and just like making it [58:30] more expensive. We'll just make it more [58:31] expensive. We'll just we'll cut out the [58:32] middleman. And so that's why everything [58:34] has gone up like to whiskey like the [58:36] cheapest of a like not anything that's [58:39] not like a Jack Daniels is like if it's [58:40] a nice whiskey, it's going to be 50 [58:42] bucks plus like because [58:44] >> that's what's happened. And it's So, is [58:46] a dinner for two anywhere? [58:49] >> It's ridiculous. [58:49] >> Okay. Yeah, that's a Yeah. So, we've had [58:52] this debate with one of our friends. [58:54] It's [laughter] like he's like, "Okay, [58:56] if you go to like, I don't know, fast [58:58] food, McDonald's, Taco Bell, whatever, [59:00] and you're like, "All right, I've got [59:02] two adults, couple kids, whatever. The [59:04] amount of that you'll spend on that, [59:06] he's saying, is actually sometimes more [59:08] than you'll spend going to like a sit [59:10] down restaurant." because like if you go [59:12] to he has certain ones you go to that [59:13] like if you get a adult entree your kids [59:16] eat free or something like that and so [59:18] it's like the kids aren't a factor in [59:19] the cost so you got two adult entre at [59:21] like 20 bucks or whatever and then like [59:25] um yeah I'm like well what about tip or [59:27] whatever and he's like no he insists [59:28] like it doesn't matter what he does it's [59:30] cheaper to sit at the like sitown [59:32] restaurant I was like that's [ __ ] up [59:33] man like that's not the way it should be [59:36] like fast food has gotten super [59:37] expensive [59:38] >> super expensive I've noticed the same [59:40] thing the Uh there's so we do Woody [59:42] Wednesdays and one of the reasons we do [59:43] Little Woodies is because it's um [59:47] locally it's like a local famous person. [59:50] It's wildly cheap, really good quality. [59:53] Their French fries are amazing and I [59:55] walk out of there and I can feed the [59:56] whole family for 45 bucks. I pretty much [59:58] can't do that anywhere [60:01] >> anywhere. Um [60:02] >> yeah, [60:04] >> it's crazy. So, yeah. Well, Seattle is [60:07] its own beast cuz when I was there, I [60:10] got Taco Bell delivered and I got uh [60:13] let's see, like a burrito and two tacos [60:15] or something like not a lot. And it was [60:18] like $67 or something like with the all [60:21] the like cuz they're like, "Hey, you're [60:22] in Seattle, bro. Like, this is going to [60:24] be at least $30 just for us to even open [60:26] the Door Dash app." And then like [60:28] [laughter] [60:29] >> like there's so many extra fees and I [60:31] was just like, "Holy shit." So, yeah. I [60:32] mean, stuff is probably just more [60:34] expensive in general. And in Europe, [60:35] >> if you buy if you're buying food for [60:37] just yourself, the order stuff is just [60:39] stupid. You're going to pay more in fees [60:41] than the food is. Is [60:42] >> that's why I always order like two or [60:44] three meals whenever I get delivery and [60:46] then you just microwave it. [laughter] [60:48] >> Oo, good idea. Dang, that's a it's a [60:51] life hack right there. [60:52] >> Yep. Yep. Yeah. And also Costco has Uber [60:56] Eats and uh Door Dash gift cards. You [60:58] can get $100 of gift cards for like 75 [61:01] or 80 bucks something. So, [61:03] >> how does that make any sense? [61:05] >> I buy them every time I go. They limit [61:07] you. You can only get like four per [61:09] month or something because they don't [61:10] want you just having infinite like [61:13] percent off at Door Dash, I guess. I [61:15] don't know. [61:15] >> This is so weird. This is [laughter] so [61:17] weird. This world is so weird right now. [61:20] >> Yep. Yep. [61:22] All right, we are at time. Is there [61:24] anything that we missed that you want to [61:26] mention before we end? [61:27] >> No, man. That was awesome. [61:29] >> Cool. All right. Thanks everyone for [61:31] listening. If you liked it, please [61:32] subscribe. Leave us some ratings. Please [61:34] actually do. I'm not going to say the [61:36] same spiel that I always do. Go to [61:38] Spotify, click our show, click rate, [61:40] give it five stars. Do the same on [61:42] Apple. It is very quick. Those of you [61:44] that are listening that haven't done it, [61:47] shame on you. I've asked you so many [61:49] times. All right, we'll catch you guys [61:51] next time. Yeah, like [61:54] you've been watching Whiskey Web and [61:56] Whatnot recorded in front of a live [61:57] studio audience. What the [ __ ] are you [61:59] talking about, Chuck? Enjoyed the show? [62:02] Subscribe. You know people don't pay [62:04] attention to these, right? [music] Head [62:06] to whiskey.fun for merch and to join our [62:09] Discord server. I'm serious. [music] [62:12] It's like 2% of people who actually [62:13] click these links. And don't forget to [62:15] leave us a fivestar review and tell your [62:18] friends [music] about the show. All [62:19] right, dude. I'm out of here. Still got [62:22] it. [62:25] >> [music]  ## 📺 Video ![](https://www.youtube.com/watch?v=9w_5Pstg1zk) --- # 📝 Full Notes (Deep-Read Analysis) # 🏷️ Title: We Accidentally Deleted Programming (Presented by CodeRabbit) ## 🧰 Action Inbox - [H1] Try CodeRabbit for AI code review — 100k open-source projects using it #ai-tools - [H2] Experiment with anchor positioning for FLIP animations — now baseline in Firefox #css - [H3] Evaluate library conversion from Svelte 5 to Solid — bundle size concerns #javascript ## 🧭 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 - [CodeRabbit](https://coderabbit.ai/) - [Whiskey Web and Whatnot Podcast](https://whiskeywebandwhatnot.fm/) ## 📝 Summary - Robbie Wagner and Adam Argyle discuss the impact of AI on programming while reviewing a Whistle Pig Liquid Death collaboration whiskey - They explore whether traditional coding practices like TypeScript and databases still matter in an AI-first development world - Discussion covers new CSS features like anchor positioning, framework bundle sizes, and the evolution of developer tools - The conversation highlights the tension between maintaining code quality and embracing AI-generated solutions - Bottom line: Programming is evolving from writing code to writing specifications and prompts for AI systems ## 📖 Main Content The hosts examine how AI is fundamentally changing software development, questioning whether traditional practices remain relevant. They discuss the emergence of "spec-driven" libraries that ship instructions for LLMs rather than actual code, the challenges of maintaining TypeScript in large codebases when AI can provide similar guardrails, and new CSS features that eliminate the need for JavaScript solutions. The conversation reveals a industry in transition, where the ability to effectively prompt AI may become more valuable than deep coding knowledge. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/9w_5Pstg1zk?t=0) — [H1] Whiskey Review: Whistle Pig Liquid Death Collaboration - Key idea: 85% wheat, 15% barley whiskey aged in coffin-shaped barrels - Evidence: "86 proof, proofed with liquid death mountain water, aged final months in 380 gallon charred American oak casket" #### ⏱️ [05:23](https://youtu.be/9w_5Pstg1zk?t=323) — [H2] AI Automation for Content Creation - Key idea: Building Swift app to automate podcast episode artwork generation - Evidence: "Pull title, show notes, transcripts, pull 5-10 topics, make them concrete things you could make an image of" #### ⏱️ [08:45](https://youtu.be/9w_5Pstg1zk?t=525) — [H1] CSS Anchor Positioning Replaces JavaScript FLIP - Key idea: New CSS features can recreate complex JavaScript animations natively - Evidence: "Anchor positioning, interpolate-size, and starting-style can do FLIP animations in pure CSS" #### ⏱️ [12:30](https://youtu.be/9w_5Pstg1zk?t=750) — [H2] Svelte 5 Bundle Size Issues - Key idea: Framework upgrades don't always improve bundle sizes for small component libraries - Evidence: "Went from 17 to 28 kilobytes, core is bigger but components are smaller, doesn't benefit with only 5 components" #### ⏱️ [15:15](https://youtu.be/9w_5Pstg1zk?t=915) — [H1] Libraries That Ship No Code - Key idea: New paradigm of shipping specs and instructions for LLMs instead of actual code - Evidence: "Ships the specs and expectations so that an LLM can do it in any language, works across Elixir, Ruby, Rust" #### ⏱️ [18:20](https://youtu.be/9w_5Pstg1zk?t=1100) — [H2] AI-First Hiring Practices - Key idea: Companies moving away from coding interviews toward prompting skills - Evidence: "Show me how you can prompt and let's see how productive you can be, I don't care how you do it" #### ⏱️ [22:10](https://youtu.be/9w_5Pstg1zk?t=1330) — [H1] TypeScript Relevance in AI Era - Key idea: Questioning whether strict typing matters when AI provides similar guardrails - Evidence: "What if you spent time investing in guardrails for AI instead of converting to TypeScript?" #### ⏱️ [28:30](https://youtu.be/9w_5Pstg1zk?t=1710) — [H2] AI Struggles with Large Codebases - Key idea: Current AI models still have significant limitations in complex, multi-file projects - Evidence: "Past four files, it starts to go downhill, even CodeQX 5.2 specialized for long-running tasks" #### ⏱️ [32:45](https://youtu.be/9w_5Pstg1zk?t=1965) — [H1] Developer Tool Evolution - Key idea: Modern AI-powered development environments offering multiple approaches - Evidence: "Cursor has agent modes, work trees, multiple agents working in parallel with visual diff comparison" #### ⏱️ [36:20](https://youtu.be/9w_5Pstg1zk?t=2180) — [H2] Multi-Model AI Development - Key idea: Using different AI models competitively for better results - Evidence: "Polyamorous with models, pit Gemini against Claude against ChatGPT, have them review each other's work" #### ⏱️ [40:15](https://youtu.be/9w_5Pstg1zk?t=2415) — [H1] MCP Server Integration - Key idea: AI tools automatically configuring and discovering the right plugins for tasks - Evidence: "MCP discovery where I ask for git stuff and it pulls GitKraken MCP to figure out how git works" #### ⏱️ [43:50](https://youtu.be/9w_5Pstg1zk?t=2630) — [H2] File System Access Concerns - Key idea: Balancing AI automation with security and control over development environment - Evidence: "Co-worker making files and has file system access, not ready for automated file access yet" ## 📚 References - [FLIP Animation Technique](https://aerotwist.com/blog/flip-your-animations/) - [CSS Anchor Positioning](https://developer.mozilla.org/en-US/docs/Web/CSS/CSS_anchor_positioning) - [Svelte 5 Documentation](https://svelte.dev/) - [OpenCode AI Development Tool](https://opencode.dev/) ## 🗣️ Paraphrase - AI is fundamentally changing how we approach software development, shifting focus from writing code to writing specifications and effective prompts - Traditional development practices like strict TypeScript typing may become less relevant as AI provides similar error prevention through different mechanisms - New CSS features are eliminating the need for JavaScript solutions in areas like animations and positioning, reducing framework dependencies - Current AI tools still struggle with large, complex codebases but are rapidly improving and may eventually handle most routine development tasks ## 📰 Newsletter **The Future of Web Development: When AI Writes the Code** This week's episode explores how AI is reshaping web development fundamentals. Key highlights include new CSS anchor positioning that can replace JavaScript animation libraries, the emergence of "spec-only" libraries that work with any programming language through AI, and the ongoing debate about whether traditional practices like TypeScript still matter in an AI-first world. For WordPress developers, this shift toward AI-assisted development could dramatically change how we approach theme and plugin development, making specification writing and effective prompting increasingly valuable skills. Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer experience: Document AI-assisted development workflows for community adoption - Education: Create workshops on effective prompting for developers transitioning to AI tools - Developer marketing: Showcase AI-enhanced productivity in development campaigns #### Community - Contributors: Develop guidelines for AI-generated code contributions and review processes - Maintainers: Establish protocols for evaluating AI tools and their impact on code quality #### WordPress - Core: Evaluate impact of AI-generated contributions on codebase quality and maintenance - Documentation: Create guides for using AI tools effectively in WordPress development - Training: Develop curriculum combining traditional development skills with AI prompting techniques ## 🧪 Hypotheses - Developers who master AI prompting will be more productive than those who rely solely on traditional coding skills - CSS anchor positioning will significantly reduce JavaScript bundle sizes in modern web applications - Spec-driven development will become the dominant paradigm for cross-language library development - TypeScript adoption will slow as AI tools provide equivalent error prevention with less overhead ### 📊 Methodology - Comparative study tracking developer productivity with AI tools vs traditional coding - Bundle size analysis measuring impact of CSS anchor positioning adoption - Survey of hiring practices examining shift toward prompting skills assessment ### ❓ Questions - How do we maintain code quality standards when AI generates increasing amounts of production code? - What skills should developers prioritize learning in an AI-first development world? - How will the role of senior developers evolve when junior developers have access to advanced AI assistance? ### 🔮 Future Research - Long-term impact of AI-generated code on software maintenance and technical debt - Effectiveness of different AI model combinations in complex development tasks - Evolution of development education and certification in AI-augmented environments ### 💡 Implications - Practice: Developers need to balance AI tool adoption with maintaining core programming competencies - Education: Curriculum must evolve to include effective AI prompting alongside traditional programming concepts - Security: AI file system access and automated code generation create new attack vectors requiring consideration - Funding: Investment in AI development tools may yield higher returns than traditional development infrastructure ### 🧠 Learning Styles Assessment Reading level estimate: 10th Grade #### 👀 Visual learners - Present: Code examples, framework comparisons, bundle size visualizations - Improve: Flowcharts showing AI development workflows, visual comparisons of traditional vs AI approaches #### 🔊 Aural learners - Present: Podcast format with detailed audio discussion - Improve: Audio summaries of key technical concepts, pronunciation guides for new terminology #### 🗣️ Verbal learners - Present: Conversational format with analogies and real-world examples - Improve: Glossary of AI development terms, story-driven explanations of complex concepts #### 🧩 Kinesthetic learners - Present: Discussion of hands-on experimentation with tools - Improve: Step-by-step tutorials for trying discussed AI tools, interactive coding exercises ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear heading structure, descriptive timestamps - Improve: Alternative text for any referenced visual elements, transcription formatting #### Contrast and readability - Present: High contrast text formatting in notes - Improve: Color-coded sections for different topics or priority levels ### 🌍 DEIB Assessment #### Cultural breadth - Present: Discussion includes various development contexts and company sizes - Improve: More diverse examples of how AI impacts developers globally #### Inclusive language - Present: Gender-neutral terms, accessible technical explanations - Improve: Plain-language summaries of complex technical concepts ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Clear logical progression from AI capabilities to development practice implications - Inductive strength: Examples from real-world tool usage support broader conclusions - Analogical strength: Effective comparisons between traditional and AI-assisted development workflows #### Tone of voice - Overall tone: Conversational and exploratory, appropriate for discussing rapidly evolving technology landscape ### 🛡️ DARVO Analysis No evidence of denial, attack, or victim/offender reversal patterns. Discussion maintains balanced perspective on AI tool benefits and limitations. ## 🎯 Conclusion The conversation reveals an industry in rapid transition, where traditional programming skills are being augmented—and in some cases replaced—by AI systems. The key takeaway is that successful developers will need to evolve their skillset to include effective AI collaboration while maintaining enough foundational knowledge to evaluate and guide AI-generated solutions. The shift toward specification-driven development suggests a future where the ability to clearly communicate requirements and constraints may become more valuable than the ability to implement solutions manually. ## 🧱 Refactor Appendix ### 📚 Literature Split Pad ### [H1] AI fundamentally transforms development workflow > [!quote] [We Accidentally Deleted Programming](https://www.youtube.com/watch?v=9w_5Pstg1zk) by [[Whiskey Web and Whatnot]] via [[YouTube]]. Published 2024. Accessed 2024-12-19. > "What if you spent time investing in guardrails for AI instead of converting to TypeScript?" Represents shift from traditional code quality tools to AI-powered alternatives. [view highlight](https://youtu.be/9w_5Pstg1zk?t=1330) #ai-development #typescript ID:rw_001 ### [H2] Spec-driven libraries eliminate language barriers > [!quote] [We Accidentally Deleted Programming](https://www.youtube.com/watch?v=9w_5Pstg1zk) > "Ships specs and expectations so LLM can do it in any language, works across Elixir, Ruby, Rust." New paradigm where libraries provide instructions rather than implementations. [view highlight](https://youtu.be/9w_5Pstg1zk?t=915) #library-design #cross-platform ID:rw_002 --- ### 🔖 Atomic Split Pad (Zettel Candidates) #### [Z1] CSS anchor positioning replaces JavaScript animations - Scope: Modern browsers with baseline support - Pointer: 08:45 timestamp discussion - Links: [[CSS-Features]] [[JavaScript-Alternatives]] #### [Z2] AI prompting skills become hiring criteria - Scope: Technology companies adapting interview processes - Pointer: 18:20 timestamp discussion - Links: [[Developer-Skills]] [[AI-Integration]] --- ### ✏️ Fleeting Queue - [F1] Test CSS anchor positioning for FLIP animations — https://youtu.be/9w_5Pstg1zk?t=525 #fleeting #css-experiments - [F2] Evaluate CodeRabbit for team code review process — https://coderabbit.ai/ #fleeting #ai-tools - [F3] Research multi-model AI development approaches — https://youtu.be/9w_5Pstg1zk?t=2180 #fleeting #ai-workflow