Live at All Things Open W/ Bdougie  ## Metadata - Author: [[Whiskey Web and Whatnot: Web Development, Neat]] - Full Title: Live at All Things Open W/ Bdougie - Category: #articles - Document Note: --- tags: [Open Source, AI Agents, DevRel] author: [MISSING] source: YouTube reference: [https://www.youtube.com/watch?v=oEeZKNJvxh8] date created: {{date:YYYY-MM-DD}} {{time:HH:mm}} --- # 🏷️ Title: Live at All Things Open w/ Bdougie ## 🧰 Action Inbox - [H1] Draft post: maintainer pipelines using agents — context/https://www.youtube.com/watch?v=oEeZKNJvxh8 #articleseed - [H3] Add “squash+rebase” section to contributor docs — context/https://www.youtube.com/watch?v=oEeZKNJvxh8 #addendum - [H4] Create AI decision log template for repos — context/https://www.youtube.com/watch?v=oEeZKNJvxh8 #xref - [H5] Prototype a TypeScript-fixer sub-agent workflow — context/https://www.youtube.com/watch?v=oEeZKNJvxh8 #articleseed - [H7] Write migration playbook: Heroku→Kubernetes lessons — context/https://www.youtube.com/watch?v=oEeZKNJvxh8 #articleseed - [H8] Ship “track-on-view” repo crawler for dashboards — context/https://www.youtube.com/watch?v=oEeZKNJvxh8 #articleseed - [H9] Build “fundamentals before prompting” study path — context/https://www.youtube.com/watch?v=oEeZKNJvxh8 #xref - [H10] Compare self-hosted AI options for WordPress orgs — context/https://www.youtube.com/watch?v=oEeZKNJvxh8 #share ## 🧭 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 - Include links cited in or relevant to the article. ## 📝 Summary - A live conversation covers AI coding, maintainers, and tooling. - Speakers stress human oversight for AI contributions. - Stakeholders include maintainers, DevRel, and regulated enterprises. - Risks include over-editing by agents and opaque decisions. - Bottom line: document choices, scope agents, and ship sustainably. ## 📖 Main Content - The discussion centers on practical AI use in software teams. - Speakers advocate “human in the loop” for open source health. - Sub-agents handle narrow tasks well, especially in enterprises. - Documentation of AI decisions preserves context for teams. - Migration stories underline collaboration over solo struggling. - Cost awareness matters with token usage and cloud runners. - Data pipelines benefit from targeted crawling strategies. - Beginners still need foundational knowledge before prompting. ### 🗒️ Notes #### ⏱️ [00:01](https://www.youtube.com/watch?v=oEeZKNJvxh8&t=1) Cold open and setup - Light banter frames an informal, candid format. ### [H1] Open Sauce legacy celebrates community signals > [!quote] [Live at All Things Open w/ Bdougie](https://www.youtube.com/watch?v=oEeZKNJvxh8) by an unnamed author via [[YouTube]]. Published [MISSING]. Accessed {{date:YYYY-MM-DD}}. > “You pushed, you merged, you shipped your legacy.” A toast marks open source impact and acquisition transitions. [view highlight](https://www.youtube.com/watch?v=oEeZKNJvxh8&t=105) #share ID:yt_105 ### [H2] Editor choices prioritize stability while streaming > [!quote] [Live at All Things Open w/ Bdougie](https://www.youtube.com/watch?v=oEeZKNJvxh8) > “It keeps bumping it back and forth…on the right side, it doesn't.” Live streaming favors fixed panes to avoid layout shifts. [view highlight](https://www.youtube.com/watch?v=oEeZKNJvxh8&t=320) #xref ID:yt_320 ### [H3] Teams prefer squash+rebase over merge commits > [!quote] [Live at All Things Open w/ Bdougie](https://www.youtube.com/watch?v=oEeZKNJvxh8) > “Merge commits are messy.” Policy reduces noise and simplifies blame reads. [view highlight](https://www.youtube.com/watch?v=oEeZKNJvxh8&t=200) #addendum ID:yt_200 ### [H4] Human oversight steers AI toward maintainable outcomes > [!quote] [Live at All Things Open w/ Bdougie](https://www.youtube.com/watch?v=oEeZKNJvxh8) > “Don't always trust the AI to do everything for you.” Review and PR comments transfer knowledge and reduce burnout. [view highlight](https://www.youtube.com/watch?v=oEeZKNJvxh8&t=1080) #articleseed ID:yt_1080 ### [H5] Document AI decisions to reduce future confusion > [!quote] [Live at All Things Open w/ Bdougie](https://www.youtube.com/watch?v=oEeZKNJvxh8) > “If you're going to make a decision with AI, write it down.” Decision logs preserve rationale and constraints. [view highlight](https://www.youtube.com/watch?v=oEeZKNJvxh8&t=1302) #xref ID:yt_1302 ### [H6] Sub-agents handle scoped code tasks effectively > [!quote] [Live at All Things Open w/ Bdougie](https://www.youtube.com/watch?v=oEeZKNJvxh8) > “You could build agents for just that thing.” Examples include code review and TypeScript fixes. [view highlight](https://www.youtube.com/watch?v=oEeZKNJvxh8&t=1163) #articleseed ID:yt_1163 ### [H7] Collaboration beats solo during infra migrations > [!quote] [Live at All Things Open w/ Bdougie](https://www.youtube.com/watch?v=oEeZKNJvxh8) > “I had to like beg engineers.” Pairing accelerates safe Heroku→Kubernetes transitions. [view highlight](https://www.youtube.com/watch?v=oEeZKNJvxh8&t=1722) #articleseed ID:yt_1722 ### [H8] Track-on-view keeps data pipelines affordable > [!quote] [Live at All Things Open w/ Bdougie](https://www.youtube.com/watch?v=oEeZKNJvxh8) > “If you ask for the repo, I will start tracking it.” Collect only data users actually inspect. [view highlight](https://www.youtube.com/watch?v=oEeZKNJvxh8&t=2906) #articleseed ID:yt_2906 ### [H9] Beginners still need core foundations > [!quote] [Live at All Things Open w/ Bdougie](https://www.youtube.com/watch?v=oEeZKNJvxh8) > “You don't know what you don't know.” Study design, infra, and data before heavy prompting. [view highlight](https://www.youtube.com/watch?v=oEeZKNJvxh8&t=3013) #xref ID:yt_3013 ### [H10] Self-hosted AI protects org data boundaries > [!quote] [Live at All Things Open w/ Bdougie](https://www.youtube.com/watch?v=oEeZKNJvxh8) > “Doesn't like sell your data off to a VC.” Self-hosting supports compliance and sovereignty. [view highlight](https://www.youtube.com/watch?v=oEeZKNJvxh8&t=3424) #share ID:yt_3424 ## 📚 References - [YouTube: Live at All Things Open w/ Bdougie](https://www.youtube.com/watch?v=oEeZKNJvxh8) - [GitHub REST API: Events](https://docs.github.com/en/rest/activity/events) - [GitHub Actions documentation](https://docs.github.com/en/actions) - [Continue.dev](https://www.continue.dev/) - [DuckDB project](https://duckdb.org/) - [Zed editor](https://zed.dev/) - If any link fails, access via the primary domain’s docs or homepage. ## 🗣️ Paraphrase - The guests discuss AI coding agents, maintainers, and workflow habits. - They recommend scoped agents, strict commit hygiene, and documented decisions. ## 📰 Newsletter - At All Things Open, Bdougie riffed on scoped agents and maintainers. - Highlights: “human in the loop,” “squash+rebase,” and decision logs. - Try sub-agents for TypeScript hygiene or PR review. - Explore Continue.dev and DuckDB for reproducible data dashboards. - WordPress folks: consider self-hosted AI for The WPCC projects. - Watch the session: https://www.youtube.com/watch?v=oEeZKNJvxh8 ## ✅ Action Items #### DevRel - Publish a “decision logs for AI” guide with repo examples. - Record a live demo of sub-agents reviewing PRs end-to-end. - Add office hours to triage contributor onboarding using agents. - Track token spend and publish monthly cost snapshots. #### OSPO - Define policy for AI-assisted changes and reviewer sign-off. - Add SBOM and provenance notes to agent workflows. - Maintain a model registry with versioning and eval notes. - Approve self-hosted endpoints for regulated repos. #### Community - Run “first-PR with an agent” pairing sessions. - Launch maintainer clinics on squash+rebase workflows. - Publi - Summary: and whatnot. I don't have a catchy name for it today. - URL: https://www.youtube.com/watch?v=oEeZKNJvxh8 and whatnot. I don't have a catchy name for it today. Uh Cabernet code and something chatter. Chatter. It's a little like Cabernet. >> Yeah, I'm I'm impressed. >> Yeah. Um Yeah. So, we are we are live here. I have my guest host, Nick Taylor. How's it going, Nick? >> Not too bad. Doing pretty good. It's been It's been a minute. It's been like what 12 hours since I last >> Yes. Seen a lot of each other. >> Yeah. Yeah. Yeah. >> And our guest today is B Ducky. How's it going? >> Uh I am super well. Well, I've actually no more talks. I'm just uh eating salty popcorn in uh in the Cabernet corner. >> Nice. Nice. Nice. Cool. Well, yeah. Before we get into anything else, let's start with some Cabernet. We got some Josh. Everyone's a fan of Josh just cuz I I think it's a silly name to just be Josh. >> Josh. I actually I'm not familiar with Josh. >> Yeah. >> I don't know about his game. >> Yeah. I always uh like brings his game. >> Yeah. Speaking of games, I will play games with my friends and like like, "Hey, Josh is gonna join us tonight to play >> right here." >> I'll pass that down. Here we go. >> I've never heard of Josh. >> No. Is this like regional? >> It must be, I guess. I didn't know it wasn't a >> No. Like, does it have a a wine? Like, what's the location of the winery? >> I don't know. Let's Let's check. >> Aside from the location in my cup. >> Nope. Got a bunch of my pants. >> Oh, Josh. >> Oh, Josh. Yeah, you're always messing with my pants. >> Sloppy Josh. >> Okay. Yeah, let's let's try some. All right. So, I've got a uh Do you have a toast for us? >> All right. Yeah. >> Here's the Open Sauce, the platform that reminded us our commits actually mattered, even if our bosses never noticed. You helped us track our GitHub glory, and now you're just another repo in the great archive of acquisitions. So today we pour one out for all the green squares and the spicy little startup that made open source feel delicious. Cheers to open sauce. You pushed, you merged, you shipped your legacy. >> That's amazing. I'm going to tear up, >> man. What a toast. Yes, that was a good time. >> Yeah, AI is good at making toast if you didn't know. >> Yeah. >> Amazing. >> Yeah. >> Yeah. Open sauce was good times. >> Yeah. Yeah. Um Yeah. Yeah. I don't know if how we're starting this podcast, but yeah, Nick worked with me at open source for a good year at that point. You were you're half the journey. >> Yeah. >> And uh yeah, now the journey is completed. I >> actually got I got word that the last person who came over in the Linux Foundation acquisition just uh announced a departure last week. So >> it it literally is finished. >> Okay. Yeah. I tried to go to it earlier and it was like no. >> It's like oh that's sad. But yeah. Yeah. Uh the end of an era. We may get a little more into that here in a little bit, but we got some hot takes which we maybe did with you before. You've been on the show before. I don't know. >> Yeah. Honestly, I still have that bottle. I haven't finished it. So, um I don't remember what we talked about, but yeah, the whiskey was good. >> Yeah. Yeah. Um so, yeah, I have a condensed list, but let's just do a few here. Get rebase or get merge. >> Um rebase. >> Okay. >> Yeah. I like, you know, pure history. Uh but you know, actually, I'm a I am a squasher. >> Yeah. >> Mhm. So squash first and then rebase and you always have historical record. Merge commits are messy. >> Oh, they are. Yes. >> Yeah. The worst is when you're actually squashing and doing things really clean but then still merging. So it's like really nice history of what you actually wanted but merged merged merged. >> It's like yeah >> I get you. Uh is HTML a programming language? >> That's that's a good question. Uh you can do cool things with it. Um, it is a template that you can write in language in natural language. >> Not even lang. It's not even natural. It's semantic. Yeah. I don't know. >> Okay. >> That's not a great answer. It's just to spark conversation. So then is CSS a programming language. >> Uh, I think it's getting there. You could do some like uh performative things in there. Uh, not just like animations, but like yeah, you can do some cool things. So yeah, I don't know. I turn complete like Yeah. I just know like zeros and ones are like no reasoning or sorry what's the turning complete? So I didn't get a CS. >> Yeah, I don't >> It's always a cool thing to say though. Yeah, something something turning complete. >> Yeah, non deterministic and uh AI. >> Yeah, I don't know. Uh >> we've run out of chat GBT tokens. Please check it back in at 10 p.m. >> You're absolutely right. >> Um okay, sidebar on the left or right in VS Code? >> Oh, yeah. Yeah. Uh, cybar on the left, agent editor thing on the right. >> Yeah. >> A right side. I don't know. Cursor is like gone back and forth. Like cursor I think a lot of cool kids are doing on the left side. Uh, but I feel like having my my file menu on the left. Like I'm just so used to that. >> Yeah. Yeah, that's the default. And yeah, so I think the only person who has said the right like taken the time to move it to not the default is Jason Langorf. He's like subjectively the correct answer. like it is not a subjective choice. And I'm like, "No, it it is." >> I I got bad news. I'm I'm one of the right side boys, too. >> Oh, are you? >> The main reason why I do it is when I'm live streaming and you're live coding, >> it it keeps bumping it back and forth if you open and close it. And if you have it on the right side, it doesn't affect it as much >> cuz all the white space. Yeah. >> Oh, yeah. Yeah. >> But I also do weird stuff like like I do stuff on purpose to hurt my brain so it makes me think, you know, kind of. >> Yeah. >> So, I don't know. You know, I might be tempted, honestly. I might actually go in the right direction. >> So, I Yeah, I I agree with that reasoning that it like it doesn't bounce your code around when you're opening and closing it. I just don't open and close it that often. So, it's like it doesn't bother me. >> I might be with you. I actually keep it closed a lot. Like unless I'm like, "Okay, what happened here again?" Like, "Let me navigate." But usually I got a mental model of like where things are and what names are >> that I don't actually open that often. >> And if you have more than like 10 files, you're just searching for them anyway. >> Yeah. Like, >> oh yeah, I'm definitely command P gang. I I >> even though I love Visual Studio Code, I'm like always using shortcuts. Like, >> we were talking about this yesterday about like people talking about Vim and stuff and shortcuts and it's like >> I don't know. I'm I basically VS Code for me at this point is just like glorified like terminal editor kind of. >> Yeah. So >> yeah, I used the uh IntelligJ keybindings unfortunately because uh I started out with WebStorm and learned all of that and I was like I don't want to change. I should have changed though because now everybody's like oh hit this thing and I'm like no my my keys don't do that. >> Yeah, it's a bit awkward. I I'm a Vim user as well and uh I got the Vim bindings but none of my key bindings in them transferred over >> but if I go to So honestly I don't even use VS Code anymore. I'm in Zed and Neoim. >> Okay. So I go back and forth cuz that I have no plugins on there so it's like fast, no bloat. VS Code opposite problem. Too much stuff in there and I don't have the time to go clean it up. >> Yeah. >> Yeah. Yeah. I've been hearing uh I think Adam Argyle, my new uh co-host is uh using Zed a good bit. He's saying it's uh come a long way cuz like when I first tried it, I don't know two, three years ago or something. When did it come out? It's been a little while. Yeah, it's been a minute. >> Yeah, it was open source like 2019. Like it was pretty early. >> Maybe it's been longer than I thought. Uh cuz I you know the history of Zad. >> It's the uh team that built Atom at GitHub. >> Yes. >> Jogging my memory. We've had the guy uh what's his name? The guy that >> I forgot his name. I can see his face. But um >> yeah, he was on the show and told us all about it, but my >> Yeah. So this kind of like had a cool project they were working on and they didn't really take funding until like 2022, like kind of later. Okay. >> Um and like it was at that point they had so many users, but then they kind of figured out what the product was going to be that would make money. >> Uh which don't ask me what it is. I have no idea how to make money. >> How does anyone make money? >> I show up every day. >> I I just ask Josh. >> Yes. >> Just look at the bottom of the cup and that's where the money comes from. >> Yeah. Yeah. >> Yeah. Josh is probably doing okay, I would think. >> Yeah. >> Yeah. Um Okay. Rails or Laravel? >> I don't know Laravel, so I guess Rails. >> All right. >> You don't have to know either. You can just have an opinion. I don't use either of Laravel seems really cool. Like they seem to be like a lot of cool kids are over there. Um I just don't have like the brain power to do another thing. >> Yeah. >> Uh so I have built a couple Rails apps recently just for kicks and giggles really. >> Yeah. >> To see what what's still there. That was my first sort of framework honestly. >> Yeah. Yeah. I think the batteries included framework model has gotten kind of lost and now everything is just a React component of some sort like cobbled together. >> I mean Nick, you're a Rails person. Like I feel like the AI world leans into like a strong opinions. >> Yeah. >> Cuz then the agent won't go off the rails, pun intended. >> Yeah. Yeah. >> But I don't know. Have you written Rails recently? >> Not recent. Like when I worked at Dev 2, uh they were a Rails shop and like you know the thing the thing I found about Rails was the to your point about the batteries included and stuff. Even though I hadn't really done any Rails until that point, it was very approachable. And uh you know I think it the thing is like you've heard this at least in the react world people have been trying to mimic what Rails does in React land. Like I I know like the remix guys they've been like trying to do that kind of stuff and Laravel I think kind of modeled themselves on like the Rails batteries included. Um >> yeah. So yeah, it's you know it's it's one of those it depends like you know it's kind of like people who like to customize literally everything like it's it's cool for a while but like at the end of the day you got to ship stuff and like making some bespoke like React combo of stuff like when we used to have like React boiler templates or stuff. It's like I don't really care that more much anymore now. I'm just like, I just need to ship something and this is good enough, you know, like >> Yeah. I think a lot of people disagree though. They don't want to ship stuff. They want to spend two years making that bespoke React framework. >> Well, how is he going to get those zero users, right? >> I know, right? >> Or or keep them. Yeah. I honestly I spent all summer building a bunch of projects and one of them actually, the one I talked about in the keynote, um, finally has users. Uh, not intentionally. Is this more I've got a weird following when people actually look at my GitHub and they're like, "Oh, I saw you ship a thing." Why are you looking at that? >> It's public because I need a free repo and I need free access to actions. >> Otherwise, it'd be private. But actions is expensive when it's a private repo, >> especially if you got to run a Mac OS one. Those things are brutal. >> Never. Funny enough, I I was at GitHub when they were uh trying to figure out the Mac OS thing and it was like the trash can Mac time >> and it was legitimately just a bunch of trash cans like the the cylinders like in a room. >> That's what it ran on. >> Yeah. Yeah, that was how it how we >> and that's how Apple does it, too. All of their stuff runs on Mac minis under desks. >> Yeah. Well, minis make way more sense. >> The trash can thing, they can't stack trash cans. >> Yeah. Yeah. The logistics are harder. >> Yeah. Well, we've come a long way. >> Those uh those Mac OS runners It cracks me up because like we've got like our product is open core so we've got a open source repo public and we have Mac runners that run and when what's the new version of Mac called Tahoe Tahoe sounds right >> it so like I was doing work and like my PR was like all the build stuff steps were passing and then >> Tahoe got released and I'm like what's going on like the builds are failing now and like that classic thing that you see on your actual Mac where it says you got to update the tools for Xcode for like get like that was happening on the Mac runners >> and I didn't clue in right away but uh so anyways >> oh that thing is so annoying >> like I just want to use git don't make me agree to something >> well nowadays like I actually get my agent so I'll use like the continuous CLI to basically set up my system so I had it set up t-mucks for me >> okay >> I've done a couple other things where my machine was like taking off all the time >> turns out I had a docker running like multiple docker images running >> that'll do And I was like, "Hey, I don't know what's going on, but could you just check to see like what's taking all the CPU usage?" I guess I could have opened up the the little uh box thing. But I was like, "You do it." And uh it works. So like I I honestly don't set up anything like my Vim my nem config was a port from my Janice Vim from back in the day. Like I had the same template, the same dot uh dot files, but I I never really updated it. Okay. >> So I was like, "Take this and convert it to Neoim." >> Mhm. Uh, and it took a couple couple tries, but yeah, it it it cleaned up my dot files for me. >> Nice. Nice. Very nice. Um, okay. Related to AI, is AI to blame when people do dumb [ \_\_ ] with it? >> Uh, honestly, it's usually I'll be honest, I give a whole talk on this, which is like it's literally the person at the end of the keyboard. >> So, like if you're bad at writing tests, you're going to write bad tests with AI. If you're bad at documentation or bad at figuring things out or even shipping code, you're probably going to ship pretty bad code with AI cuz like when you know what you don't know, like you do a better job. >> But if you're just like, oh, I'm going to reimplement um sorry, I heard a story of a of a company that a lot of people know about uh and there was an engineer was like re-implementing a a programming language >> uh using AI credits that they had at the company. Uh it was pretty fascinating. um they don't work there anymore. But like when you just do like dumb things like that, like why would you need a new program language for like this problem? It's like AI can like build something spoke for this situation. Like it's probably not the the right tool for the job. >> Yeah. Yeah. I feel like uh I don't know there need to be some kind of disclaimers maybe because like people don't think about what they're doing and like AI just wants to solve the problem. It's not caring if it's like secure or good. It's just like, >> yeah, all right, I'm gonna expose all your secret keys because I need to like push them up to this like thing to do something. >> The script that has like literally the service token directly in the script. >> I I've had the thing where um well, I've got that, but I I've created rules to like never do that again. >> But the one thing is like when you have like 20 tests fail because of some legitimate change and like got to update the test, >> it like AI doesn't really know. Well, at least I I'll pick on cloud. So the clawed sonnet model doesn't know to like refactor. It just knows to add. So it always adds green code around the place. >> Okay. >> So then it'll be like, okay, 20 tests fail, 17 are passing. We're good. Let's move along. The three are not don't worry about those. >> Let's just like move forward. I'm like, dude, what are you what are you talking about? Like fix the test. >> So I have a rule that's basically if you touch it, make it better. >> And it's actually pretty good because it will actually look at that. It's like the first one up. If you like if you touch it, make it better. So if it's like, oh, a 17 test pass, well, if you touch it, make it better. So fix test three test. >> Yeah. >> It's like, oh, you're absolutely right. >> Get the classic. You're absolutely right. Yeah. >> Yeah. Great answer. I actually had Codex come back and be like, uh, man, it was like uh Oh, it was like it was like OMG or something like that. It was something dumb because it tried to be cool. Yeah. >> So I was like, "OMG, I can't believe I did that." It's like what the please stop. >> It's like that Steve Bushami with the skateboard. Hello fellow kids. >> Yeah, that's GBT5 right now. Yeah, it does. It's It's like uh Garsh Darn or whatever. Like, dude, you're trying too hard. Just like ship the code, dude. >> Yeah, >> put the fries in the bag. >> As long as it's not saying 67. That's all like my >> Oh, boy. Oh, man. >> Yeah. All right. So, we're at All Things Open. >> Tell me about how open source has influenced your life. >> Man, open source, uh, I don't think I'd be here without open source. So, back when I was just a struggling artist. No, I'm just kidding. Um, no, actually, I got I actually mentioned in the keynote, my intro into programming was always copy and pasting from stuff I could find in open source. >> Sure. So, so I used to play a lot of like uh Counterstrike or um uh Rainbow 6 Rogue Spear, like you would mod it and find something on Source Forge and I was like, "Oh, cool." Like copy paste, do something. Or back when action script was a thing with flash games, like, "Oh, copy and paste, do a thing." Or MySpace, copy and paste, do a thing. >> Uh so when I decided I was actually going to like try to build something for real, >> uh I ended up just like taking uh existing code and remixing it into like essentially Hey, what's up? uh remixing code into like making it work for me. Yeah. Uh which I didn't know that was like that was a real thing. People did it all the time like Stack Overflow development. >> Uh the that's how I learned was like oh solve problem copy and paste and shift it into like how I can make it >> which is perfect for AI. >> Yeah. >> Like I'm 100% like oh >> shape of what I want to do read this file build it again for me but make it match my design system. >> Yeah. >> Uh so like it is it unlocked uh where I have a CS degree. I was able to like just get in and feel like I could do something. Uh, and that's how I've been >> managing my imposter syndrome ever since. >> Nice. Nice. Yeah. So, you mentioned your talk a few times. You want to give us the the highlights and what you talk about? >> I gave two talks. So, one was about finding future maintainers in the in the world of AI focus around like a AI agents are like they do a really good job of doing things, but that's not always the right thing. Uh but we have got this bit of like it's always an epidemic around open source maintainer like people being burnt out. >> Yeah. >> Like there's not a lot of people left over uh to actually write the code. Yeah. >> And it's even worse now with AI. >> Uh so I was like encouraging folks to like write issues, do PR review, like introduce the human in the loop. >> Don't always trust the AI to do everything for you. Uh because then that's the knowledge transfer. Uh so like once the PRs merge the stories ends and like that's at the point where the story starts where you can like go back and be like hey agent let's go read some PRs together and I told the story of like when I joined GitHub day one you would sit in a room with all the people on boarding uh and they would teach you how to use GitHub for 90 minutes and I had already been like coding full-time at that point. So most other engineers are like just doing a bunch of stuff. I was going to the PRs and like going all the way back and pageionating and then looking at all the all the founders code >> and looking at the issues that were open like all the incenderate things that I knew about outside. I was like I looked for those issues. >> Yeah. >> Um and it's like crazy how easy it is like to see that when you work at GitHub cuz it's all in the repo like nothing's deleted. >> Um I don't think when Microsoft acquired us like if they haven't deleted anything it's just like there in the open uh full full value. So, um, yeah, like storytelling through code like but don't forget to like write docs. If you're going to vibe code the product, at least write some docs so the vibe coding can still happen the next time you're in a session. >> Yeah. >> Uh, and then this afternoon I get a 45minute keynote on continuous AI. So, I work at continuous my day job. I do DX there and the focus is you can build u custom agents for bespoke situations. So like if you have an agent that just code reviews or have an agent that only fix TypeScript errors and like make sure your TypeScript files are pristine like you could build agents for just that thing. >> So whether you like you like AR or you don't like there are things you don't like doing like I don't like writing types. >> Like I honestly I would be the any person like the way AI is like oh I'll put any there or unknown. Yeah >> is like that's me. Like you're me and the AI are like we're synergistic. Um but I got to force the AI to be like no let's write some types. Uh, I have rules around types. This is how we reuse types around the project. Um, so I gave like a 45minute demo that actually kind of failed. Um, because we actually shipped some stuff that broke broke the build. So I had to do some handwaving on our hosted thing, but I showed a bunch of like local CI CLI stuff. >> Nice. Nice. Yeah, I hadn't considered like making specific agents to be like you just care about types. Yeah, it it is actually very rewarding when you like get very granular on that where like there's no one at your company that's going to be like, "Oh, I'm the typecript person." >> Like you're probably going to get fired. Like dude, like if you spend all your day like massaging types and reviewing types, like that's not going to happen. But like for an agent, oh, why don't we just have a type >> Typescript agent? And this is how like when I worked at GitHub, they had specific knowledge people like we had the GraphQL internal GraphQL team. like that was not their full-time job, but they were like the the torchbearers of like how GraphQL worked at GitHub or like the payments team, which they didn't have a payments team, but there were some folks would go other places, but come back for code review. Um, and there like certain individuals when it came to like system architecture and database design where like when they open an issue like everyone stops and reads it, okay? >> Cuz like this is the future of GitHub. Like when we shipped GitHub actions like there was an architecture design a long docu uh and a couple issues and a couple demos that were recorded of like this is what we're building like this is the future. Same co-pilot like uh when UG u is a CEO of expo he left um to go do a new company but he also was the first uh it's funny there's a lot of people who claimed to be like the creator of co-pilot or the folks who named it like that's all true. there's like a a team of like six to seven six seven um but UG was the one that kicked off the idea of like taking Openai and doing a thing with it >> and um so like when you started doing demos they were pretty bad demos but they were like really interesting and like there was like interesting story there that you can like stop and listen and learn. So like uh we're losing a bit of that in the AI revolution that you should like if you're going to make a decision with AI, write it down. So at least the next person is not going to be very confused on like why this code is like weird. >> Yeah. >> Yeah. >> It's a good point. >> I I feel that I I was curious uh you mentioned Well, I know you work at continue, but you're talking about the agents. So does that mean you guys I know you shipped uh agent runner for like CI/CD pipelines. when you're talking about the agents like the TypeScript agents and stuff, are those sub agents kind of like Claude or >> Yeah. So, honestly, that's that's actually the best way to kind of mental frame this is sub agents and claude. Um, but sub agents and claude that you can host and share across your team. So, like with claude, you can host those specifically in your claude >> uh folder in your repo. Uh, but what I love about sub agents and claude is like they're bespoke but also ephemeral. So if I I I have a refactoring agent in cloud like I'll delete that all the time and like things change all the time. So like let's just delete that and with continue it's less about deletion but it's more about sharing. So if you have a team like we talk we actually work with a lot of enterprises across multiple teams like most of our enterprise sales is like 100 minimum 100 seats. Uh so you're pro you're trying you're platform engineering trying to scale AI within an organization. there's a chance that you figured some stuff out that you want to go into the room and start training other engineers. So like you build all these sort of sub agents or we call them agent workflows uh to basically say okay this is the TypeScript experience. Uh this is if you ever have a typescript or a JavaScript thing, this is a thing you pull for. Uh and then you could also make a choice of like oh we can actually go like uh what is it a cl like not sonnet but ha coup 3.5 >> cuz like it's cheaper. It's so much cheaper when you have rules and a bunch of other stuff and prompts set up for your agent be successful where you don't have to use the biggest model all the time. >> And I think that's our difference between like quad cloud selling models like they're literally a foundational model. So they're always going to push you to the biggest model. Uh for like but in enterprise you might not have access to all that. So like we can get you getting that same experience on a smaller model. >> Okay. >> Yeah. And you support all kind like you so you support like open AI anthropic and um do you do local models too? Like I remember >> local models is how we started. So uh our CTO came out of MIT. Uh he actually went to school with the CEO of cursor. Uh so graduated same class. Um and so but he built the VS code extension to make Oama work in VS Code. >> Uh so this predates co-pilot predates almost everyone. Um and that was like the focus for like 18 months. Uh and then early this year we built a SAS product. Uh to essentially like uh you could also bring your own tokens in the the local version. Uh but the SAS product basically gives you a hosted experience. So that's the SAS product that enterprises use the hub hub.ontinu.dev. Uh so like if you want to deploy this to your your thousand engineers across your your global team distributed team like this would be the product you sort of have that data plane orchestration on. >> Okay. So I'm not sure I fully understand. And so like, >> oh yeah, buzz word >> is it? Um, >> it's like if if you have a team of engineers >> and they all want to do a certain task like the TypeScript thing you were talking about like they would log into this hub, choose the TypeScript one and be like I want to do this thing in that or like >> Yeah, more than likely you're not even choosing like it get kicked off. So unfortunately my demo failed that I didn't get the part in the code review >> but like code rabbit and kodo are both sponsors here and like you could just like install those code review tools or use copilot off like face value but the interesting part about that is like where's that data go >> like if the hosted server your entire codebase goes off in the like the it's hard to use enterprise basically so like >> uh I have a code review agent that basically I have rules on how my code is written in the project and how we write documentation so I will have a code review agent workflow that kicks off every time the PR opens. >> So, it's funny cuz I didn't get to this part, but my entire talk was going to talk about the GitHub uh action workflow that this kicks off the agent and the way we power this is through our CLI. So, just like cloud code a CLI, but our CLI is that not just meant to like deliver for the agent, give an experience, it's meant to literally just give you a custom bespoke config. Uh so then you have like rules for your specific situation. So I I'll land the plane with in Germany they have car manufacturers that use continue and the reason why it's important for them is they have specific bespoke language that runs on your computer in your Mercedes. Oh sorry uh but that runs on your computer in your in your your luxury luxury car >> of unknown origins. >> Yeah, whatever car who knows what German luxury brands there are. But basically those languages that car manufacturers use have to be me like no memory. So it's like a language that's not trained on cursor. It's not trained in copilot. It's bespoke. Has to run in your own models, your own training. >> Okay, >> we work with that situation. >> So if you're like, okay, memory safe, like all the stuff that's like very specific to like cars need to run 100 miles or an hour or 100 km an hour and not crash, like you cannot have a stack overflow while you're trying to pump the brakes. >> Uh so there's like regulatory places and like finance and stuff like that where continue works very well out the box because we're open source for first. Uh, so like think about having your custom cursor batteries included experience, but you ship it for your team within the enterprise. >> Okay, gotcha. Gotcha. >> That's it's it's it's hard to gro if you don't work at a large company that has a bunch of these like >> best practices or like think like rules and regulations. >> Uh, but it works very well in places that are highly regulated. >> Okay. Yeah. Interesting. I don't know. I'm never allowed to use any fun things. I work for IBM, so it's like >> Yeah. Hey, >> IBM amuses us. >> Oh, do they? >> Yeah. Yeah, we just >> Well, maybe I can do fun things. >> We did a co-aunch with um Yeah, he's talked to Gabe Greenberg. He's the He's like the Devril architect guy. >> I'm Hashi Corp. So, we just got acquired. >> Oh, you're on the other side. >> Yeah. So, I don't know like we're having to change everything we do to like everything they do. And >> I experienced that with the We were at a small startup inside of Microsoft. It's called GitHub. And uh we GitHub still uses Slack to this day. And I imagine because GitHub's all the chat apps and the deploys happen in Slack. >> You can't do that in Teams. >> Same same with us. Yeah. Like we had to switch from Zoom to Teams for meetings but like still chat in Slack and do all the stuff there. >> Yeah. >> Yeah. Yeah. But what I love is like so we the the the Terraform MCP like if you got to like spin up like an environment or do a thing that's ephemeral like what's great about this like you could hook the MCP up in the continue to your models and run this like deployed on the cloud without needing to have a lot of DevOps behind the scenes experience. One of the hardest things I ever did was when we migrated off of Heroku out of GitHub >> like I had no knowledge of DevOps but we had to all move to Kubernetes >> and I ran this your your sweater was stars.github.com github.com. >> Oh yeah. Yeah. >> I had to migrate that that site off of Heroku cuz Heroku had a vulnerability >> that we found. Uh so like before we could disclose that we had to move all of our stuff off Heroku for security reasons. >> Oh yeah. >> Okay. Okay. >> And I didn't know how to use Kubernetes. >> So I had to like I had to go beg engineers because I was in Devril. So I had like beg engineers to be like hey I know this is not your job but like I have no idea what I'm doing and these docs are like they're cooked. Like I have no idea what this is. Um, so like and I I think I use cooked right? I might have used that backwards. >> We'll ask Chad. >> Yeah. Hey chat, did I use cook right? >> Yeah. Are we cooking? >> Yeah. So basically I had to like beg an engineer to be like, "Hey, can we like pair on this?" >> Uh, but it would have been great to have an agent and be like, "Hey, check out we call it MODA. If you can like the Moda MCP like in conjunction to like your Terraform thing because we also leverage a lot of that stuff." Uh, can you just like get me unblocked and like deploy my thing and make it work? Um, so it might have took me like I think it was like a uh it was like around Christmas time. I might have spent like months uh like January to March like trying to figure out myself and then finally like shout out to Mike Mcuade uh who's like the co-founder of uh WorkBrew uh also maintainer of homebrew. >> Yeah. >> He just did it for me. I'm like dude thanks so much. Like >> I don't know what I'm doing here. Like I'm just like I just stand on stage and talk about cool things like Beyonce. um by >> and shout out to all the people that have helped all of us with all the things cuz that's part of like being a senior engineer I feel like is knowing the people that know the answers like >> find or knowing the way to find the answers that's the people the sources the everywhere you might want to look but like yeah just trying to do it yourself is like you're going to waste a lot of time. >> Yeah. And it's like man in a remote company it's even worse. It's like perpetuated to like you're like, man, I want to like DM people on Slack and be like, "Hi, >> let me disengage cuz I'm too scared to say I'm dumb." >> Like, hey, hi, I'm dumb. I do not know how DevOps works at this company. >> Uh, don't tell anybody. But also, can you do this for me? >> Like, yeah, there goes your performance review. >> Yeah. >> Yeah. All right. Who needs some more Josh? Anybody? >> Uh, I can take another Josh. >> I'll say hi to Josh. All right. >> Shout out to the Render ATL popcorn that's downstairs. Um >> Oh, branded popcorn. >> Yeah. Yeah, they had they do have popcorn downstairs and it was um >> Yeah, it was quite salty. I don't know if I got like the bottom of the batch, but man. >> Okay. >> It was good though. I I needed it. I didn't do lunch. I missed the uh the salads and uh sandwiches. >> Yeah. I uh remembered about lunch when they were like pushing all of the lunch away on carts as fast as they could. I was like, I'm going to just skip that. >> Yeah. Yeah. I don't know if this is like a uh are these team stores, but they're like on the money on this uh this venue. >> Yeah. >> Yeah. They're like cleaning up stuff, putting stuff away like, "Hey, I wanted that sticker. Like, where's that going?" >> Yeah. I I got to say though, shout out to Todd Lewis and the whole crew here cuz this is my second I was at the AI one too in March, but like they run a tight ship. Like it's it's really well executed. Like >> I got into I give my talk just before this and the AV people just came in. They're like, "Boom, set up lights." They're like I said, uh, there's something up with the projector. They just went boom boom boom and I was go time. >> Yeah. Yeah. Shout out to light mode too as well. Uh, I love this this community cuz I I tell everyone before my talk like, "Hey, interrupt me. Tell me like ask me questions. This is your time. You could have like literally bounced and like did something else cooler than listening to AI talk." Uh, but yeah. Uh, uh, one of the the attendees actually shouted out like, "Hey, switch to light mode." Cuz the uh, the room was so bright. I didn't realize how bright it would be. >> Um and I had to figure out how to change the light mode cuz like who uses that anymore? >> Yeah. Like >> Yeah. I don't I mean I I set everything to dark. Sometimes I'll like do it based on the time of day, but like usually dark. >> Yeah. I used to do that for uh GitHub. I would switch to light mode. Okay. Um >> in the daytime and then dark mode. I actually had uh I forgot what plugin it was, but it would change your your system uh dark mode to light mode based on the time of day. Yeah. >> Uh I don't think it actually works anymore. It's like too many versions of Mac OS. >> Yeah. A lot of private APIs there probably. >> But >> you know, I'm going to vibe code that. >> Yeah. >> Yeah. >> Yeah. And Apple script. >> Oh man. Yeah. I >> I did a thing one time. Uh you run boot camp before I assume on a Mac where you can run Windows. >> So you know in on the Windows side you can just click the little thing and like restart into Mac OS easily. Mhm. >> But on Mac OS, it's like you got to go into settings, you got to change your startup disc, you got to press all these buttons, you got to put in your password, you got to like it won't just let you restart. So, I wrote a little like automation script thing that like clicked all that and just put a thing in the menu bar to do it, but just stopped working after like a day. It did not Yeah. Very unreliable. >> Um, >> yeah, that's how I write software. Unreliable. >> Yeah. Consistently unreliable. >> Yeah. Yeah. >> Oh, man. >> All right. Um, so you tweeted that I'm convinced there will not be a clear winner in the new AI coding era. Just a lot of happy devs getting new toys all the time. I'm curious what makes you say that you don't think someone will win and uh, everyone else will fall by the wayside. >> No, I think who takes the most money will win. Like I I think like OpenAI is like what a bazillion dollar company and like cursor took a $900 million round. Uh, but that doesn't promise people to be like actual winners. Uh, it just gets a lot of momentum into their camp. Sure. >> Um, but I actually tweeted that because I think uh it was like the week that GitHub announced a bunch of stuff and then OpenAI announced the codeex model for their u their codeex product as well and then I think Claude 45 uh Sonic 45 shipped uh and then a bunch of other stuff like it was a week before the Granite 4 models launched. So, like when it comes to models and tools, I think we're like we're good. Like there's so many cool things around like AI like if you don't want to do it in the command line, there's like Sonnet. I'm sorry, not Sonnet. There's um >> was it Sculptor came out a couple weeks ago and then there's also the conductor. >> Sounds fancy. >> I haven't heard of Sculptor. >> Yeah, Sculptor is like extremely new. It's a company that like raised $200 million to train models and then their like entry point into the market is just like really like nice experience when it comes to like UIs for like interacting with cloud code. Okay. >> Um, >> but then like like all the tools are out there like we could just leverage them and a lot of them are shock like you got to pay for some stuff but like a lot of it's like shockingly cheap. So like >> I pay for Codex. I think >> they can't stay that way though. >> Yeah. Yeah. I mean well at the Google Open AAI Microsoft level like 20 bucks a month is fine >> but for companies that got to make real money probably not. Like Devon's like 500 bucks a month. Uh I've never used it but it's it's enough that I won't use it. I just don't like the concept. Like the whole branding around it was like [ \_\_ ] you. >> Yeah. >> Yeah. It's it's real because I don't know, man. I interviewed for that company and like it felt like they didn't know what they were doing. They just kind of like they had a really good moment that they had momentum from. So like shout out to team >> now. They have Windfur, right? They're the ones that acquired Windsurf. Yeah. >> Switch is over there too now. >> Yeah, Switch is there. They I think Switch is excited about what they're shipping soon. So like we'll see. But at the end of the day, it's like I'll eventually use it. Like I was never a cursor user until I joined continue because I'm like I should probably use a bunch of other stuff. >> I was very much like a chat with chat with co-pilot >> and like don't edit my code. Just like tell me stuff and I'll write it. >> Yeah. >> Uh and I I was perfectly fine with that. But then I got a job in the space and I'm like oh okay let's go all in. Let's see how far it can take me. >> Well it's annoying when it's doing it for you but doing a bad job because it does a lot of that and I'm like just stop please. like give me a little suggestion and then we'll iterate from there instead of like oh I just edited 100 files like >> oh man I had like um I I set up a pre-commit hook cuz I I ended up starting a project contrib.info info. I started as bolt new. It was vibe code. I was like, "Hey, let me see if I can like repeat some some things that I like I I want in open source that has now been sunseted, but I'm like, I just did that chart, that one chart with the faces. Can we make that happen?" >> Oh, you still on that one? >> Yeah. I just dropped a uh a Figma attempt, like basically a Figma design, and it made it like cool. >> I get this AI stuff. >> Yeah. >> Uh but then it had like any types everywhere, so it's just all really bad Typescript code. Uh, so then I'm like, "Cool, let me just like set up a llinter so I can start fixing the stuff." 3,500 linting errors. So I'm like, "Okay, let me set up a rule that you don't fix any linting stuff unless you touched a file." >> So like if you touch it, fix it. >> Um, but every now and then it will forget >> and it will like touch 300 plus files cuz GitHub now doesn't tell you how many files it touch. It's 300 plus, >> but it will touch every file and fix it. I'm like, "Uh, no thank you." like thank you for so much for for fixing this but we're not doing this right now. So like if we touch it we fix it. Don't go off and like feel like you're going to like change the earth. And that's like 100% junior engineer like I would do this thing and be like oh I converted our app to Typescript and be like did we need this? >> No, but I read a blog post. >> Yeah. >> Yeah. Yeah. >> Yeah. Remember those days? Like I read a I went to a conference talk and then I rewrote our code into whatever random thing that someone said and like this will prove it. We're good. >> Yeah. >> Yeah. That's that's AI. >> Too enthusiastic. Yeah. >> Yeah. I uh I had a hard time with cursor because I don't remember what I was trying to do or like exactly what happened, but the gist of it was I was trying to edit this file and it was like here's what you need. Here's what I'm going to do. Like I'm going to do this stuff. I'm going to edit it like this. I'm like cool do it. It's like done. And I was like, I can I'm looking at the diff. You You've made zero changes. Can you do it? And it's like, yeah, yeah, sorry. My bad. I'll do it. And I was like, it did no changes again. I was like, >> oh man, comedy. >> Like, are you just going to keep telling me that you're going to do the changes and then not doing? It's like, yeah, that's what I'm doing. I was like, all right, cool. Let me know. >> You got on gas mode. It's just gassing you up. It's like, yeah, I got that. I got that. I know what I'm doing. >> Yeah. >> Like, no, you don't. This is uh my experience with Gemini CLI which is uh I I feel like I like this is my second glass now or cup. >> Josh is coming through. >> Yeah. >> Yeah. Josh, thank you so much for giving me the confidence to share these stories. >> But Gemini has this router built in and this is the world that AI is going to where you don't really choose your models anymore. It just routes to the best model. Perplexity has this as well. Most of the like chat GBT will route now. Uh but Gemini is like you pay $20 a month Gemini Flash. It's like, "Oh, yeah, yeah, good idea. Gemini Flash, like smallest model." And I was like, "Hey, let's like rewrite some stuff." And it's like, "Cool." It like wrote one JavaScript file. I'm like, "Hey, the app's TypeScript." It's like, "What's going on here, man?" >> Yeah. But it's like it's it's like I know what I'm doing. Stop stop asking me questions. Here we go. Here's a type here's a JavaScript file. And yeah, it's it's actually comical and how how bad it can be sometimes, but like there's ways to make it better. like add rules, add context, make it read a file before you start writing code. Like don't go to auto like vibe coding mode. >> Like interact with it first before you go into like auto accept changes. >> Yeah. So I've learned a lot of that stuff in the last couple months and it's it's been rewarding at least. >> I guess the frustration is like when you're really deep, you're starting to feel like it's a real person talking to you. So you're like, "Hey, you sound like you know what you're doing. Why don't you do it?" And it won't do it. And it's like I forget that you actually don't know anything at all. >> Yeah. Yeah. It's a reflection of yourself. You're looking at the man in the mirror. It's like the AI is just reflecting back to what you are. >> Yeah. >> But I wish it would be more humble. Be like, "Hey, actually, I don't know anything, but like based on the info you've given me, this might maybe kind of work." Like then I'd be like, "All right, that's acceptable." >> Well, you know what's fun is actually take like a code rabbit review and then tell like the quad bottle be like, "Hey, critically review this review." And it'll be like, "This model made a wrong decision. I think it hallucinated." And like it just becomes super real when it's like not them. Yeah. >> It's just like, "Yeah, this is dumb." Like, "Don't listen to this. Only take these changes." >> Yeah. >> And I love doing that because I'm like, "Now comment back on the repo and at mention the review bot." And then the review bot like, "No, >> they're wrong." >> And it'll just be like back and forth. And if you have enough of these review bots in there, it will at mention each other. And like it's it's a back and forth that will go off. So, it's like one of those things where you like take the two phones. Yeah. >> And you talk talk them to actually phones are like this now. Yeah. I was like, "No, this is >> a rotary device that it's like these two phones and you like let it talk to each other." >> Like it it gets a little fun in the review bots. I wish I I'll probably do a demo on this. I'll make a video about it. It's a lot of fun. >> That seems fun to watch. >> It is fun because like if you if you tell Claude uh cuz I use Claude for like um and and uh GitHub. >> Uh but if you're like, "Hey Claude, GitHub reviewed the code like this." in the cloud like >> amateur >> why do you keep using that tool like it doesn't know what it's talking about like this is how I would actually instruct the code ignore this stuff only change this yeah >> like hey GitHub uh said >> Claude said that you're >> the gossip >> you're you're a wussy >> I don't know you cursed before I don't know >> you can say whatever you want yeah this is not a family show >> I can say it said you're a [ \_\_ ] Oh man. Yeah. Yeah. Quad said you don't have the coahonus to actually review the code because you don't know what you're talking about and then it finds your quad token and deletes it. I see what you how do you like them apples. >> Yeah. Actually, it's funny enough. I I use G uh GPT in my app and every now and then the quad model will be like, "Oh, I've actually rewritten to use quad." And I'm like, "No, no, no, no, no, no. Stop what you're doing right now. We use GPT for inapp browsing stuff. We use the cloud model to write the code and it's just like it every now and get a little overzealous like oh hey I wrote this uh plan up but um instead of using GPT040 like we're going to use claude sonnet for this right? It's like no get out of here with that. >> Oh I love it though. Claude Claude's literally a dev rail AI agent. >> It's constantly selling you the stuff you should be using. >> Yeah. It's so funny too when like like because I I use it to I don't just say make a blog post for me. I bring you I pull in sources stuff I've been working on. I'll tweak it and claude and be like this ready to ship and then I start a new conversation. I paste it in review this critically and then it's like and it's like >> man the phone things again. >> Yeah. But it gives you the confidence like it's like a quad just a it's a developer of a couple beers in them. It's like hey I'm good go all night. >> Always on the bomber peak. Yeah. Yeah. All night until you run out of tokens. >> I guess that's the ultimate goal is to just string you along until you spend all your money. >> Yeah. >> Yeah. That was that was the uh the case for like when folks were doing quad code early days where you're like, "Oh, don't pay the $100 or whatever it is. Uh let's just use tokens." And like quickly you're like, "Oh, I just wrote a commit message and I paid $14. Like what happened?" >> Yeah. I was using uh Klein, I think that's what it's called, right? And it's like that where it's like, oh, each thing you're doing is like, you know, a cent or whatever. But then, you know, $25 later you're like, oh, okay, let's stop doing this. >> This is good. This is like a buying the first round of drinks. Like, hey, how many people are here? >> Like, how do we get to this number? >> Yeah. Yeah. >> Yeah. That's like there's uh I think I was mentioning a few weeks ago or or when we when I was in SF last time, there's um there's this project called sniffly.dev, dev which works well with cloud code and it's like >> I'm pretty new to Kubernetes still and like my talk today I was talking about Kubernetes and I had to get like a cluster up and stuff >> and like my head of engineering is like yeah he has alerts on spend for for and he's like yeah I can tell when Nick's like working on a demo because it's like it went it goes from like $4 a day to like 180 bucks or whatever but it's uh but but the funny thing is like that sounds super expensive And but but at the same time it's like I spent like maybe a day day and a half like just kind of troubleshoot something and then I didn't bug anybody. So like somebody's salary like 180 bucks is like a drop in the pond compared to like >> Hey, speak for yourself. >> Yeah, I'm just kidding. >> Yeah, you know that that's true. Uh we actually have a uh we actually track how our engineers and how much they spend on models and tokens because it actually helps influence how we plan our product and how folks will use it. >> Uh but I think I was like rewriting the docs and um I'll use like the cloud actually using opus to be like because opus is great for the context window and like deep thinking. So if you're like hey look at all these PRs let's actually let's scaffold out some docs and I think I did like a Monday and it was like $400 for the day. >> Oh wow. And I was like, "Oh man, amazing. Let's keep going. >> Spin, spin, spin, baby. VC money." But yeah, but it's it's good to see. I like when I see those numbers and like when I run a PR and it tells me like, "Oh, you spent 13 cents." Oh, okay. Cool. That's 13 cents of my time I didn't have to like spend going back and forth fixing types. >> Yep. Yeah. Totally worth it. As long as you're keeping an eye on it. >> Yeah. Yeah. So, going back to my tweet, like this is literally like we're in the heyday where I spent a ton of time not writing code uh and doing the whole Devril thing where I'd like ship a bunch of tech debt and like a bunch of side projects. I think I had like one PR inside of GitHub my entire 5 years there >> cuz I knew like my code I'm not maintaining that and that like promise of like if you're going to open a PR we're going to ask you about it later. Yeah. >> And it's like I don't want that I don't want that job. I didn't sign up for that. Uh, but nowadays I'm like, "Oh, I I I spent all summer basically doing Python and data science because I was super interested in like learning it, but I'm not going back to school for that. Like, sure, I'm not gonna bug my friend to like do it for me." >> Uh, so I was just like, "Cool." U, so I built a thing that's like DTHub uh, and DuckDB, which are just Python data science tools. Um, but I built a sandbox to basically work inside of GitHub repos and like just start asking questions. It's actually what John built for us in the SQL editor thing. And uh that's how I do like feature development. It's like just getting SQL and like think about like what data I can show like what charts I can put together based on the data I have at hand. Uh so I just rebuilt it and I was just like I know how to architect this thing. So like I just got the quad desktop uh it spit out a bunch of things. I might have seven times like built out a plan for this and then I finally was happy with like where it got to and how I would maintain it. Uh and I've been like running this thing uh for the past 3 months it's been running and on a fly box. That's awesome. >> In production. Yeah, >> that's for contributor.info, right? >> Yeah. Contributor info is like literally uh we called it pizza time uh back in the open source time. So, uh pizza time was like our our data pipeline is like rebuilding the GitHub archive. Uh I essentially rebuilt that um just through ideas and chatting with with AI. Um cuz I wasn't the person that built it, but I knew like the concept. >> Yeah. >> And I I had a concept of a plan and uh I took that plan and like implemented it, which is Yeah. I feel like I'm in a a bit of a renaissance when it comes to AI and like getting back into my uh getting my groove back essentially. >> So So you drinking from the fire hose the the events uh API from Okay. Okay. >> Yeah. Yeah. So the events API on GitHub you go to api.com/events you get at the time I open source it was every seven minutes you get like new events. Uh they've switched to every hour. Uh but you can just pull that and get every event every repo you want. So I don't get every single repo. Okay. Uh, I only get like 51 repos. So, if you're in contributor info, you can opt in, say, "Hey, I want this repo to be tracked." Uh, and like right now it's 53 or 52. And I'll just like run it on a daily crawl to like go catch it up. >> Uh, and I have like a clever thing where you're like, >> uh, so this is like the problem with open source. It got very expensive because we have 300,000 repos, >> but people are looking up maybe like a hundred of them. >> So instead, I've inversed that where I will if you ask for the repo, I will start tracking it. If you never view the repo, I will delete it. >> Uh, so it's like page view based data fetching. >> Uh, so like if you >> So you just set up a bot to keep viewing it and then >> Yeah. So basically I set up some ingest and some superbase edge functions. Okay. >> Uh, that type right into my Superbase. Uh, and then basically like you're I keep logs of like who's viewing what. Uh, and so like all the repos I care about, the four of them that I look at all the time. Uh, it just stays always fresh. Uh, and then stuff like Microsoft VS Code, like no one's looking at that. like it was just like a thing I showed once like it deletes it. >> Yeah. Yeah. >> Okay. >> So, I want to regress back a second to you were saying uh like uh you were learning Python and doing some data science stuff whatever. Um you as a established engineer had success with that. I wonder what you think about people that are new to programming, want to get into the industry. Is AI hamstringing them because like they don't actually learn it right or like what do you think? Man, you say hamstring, but like I've got this Achilles thing that's happening right now because I play basketball twice a week. Uh, so I've been like doing these stretches to like I don't want my Achilles deterior like a Jason Tatum. >> I'm not out for the se I can't be out for the season. I got to travel. Um, but yes, so like if you're coming out of college and you're like, man, how am I going to get a job a boot camp grad? Like God bless you. Um but like my recommendation for a lot of folks I was in Nigeria did a workshop is like you got to read a book and it's like the inverse of like I could learn how to code in 13 weeks or whatever it was six weeks. Um now you actually have to dig in to figure out like why am I doing this and like when >> Mhm. >> and like before you like spend all day every day like reading books and learning about languages and never implementing things. But the reason I would say like read a book or catch up through that through uh that type of stuff is because you don't know what you don't know. So, if you're like prompting build app to do like whatever, uh my recommendation is like you don't know if that's the wrong thing because you don't have the experience, >> right? >> And the only way to get the experience is by either building or asking other folks that know it or they're reading books. >> Yeah. >> And a lot of folks won't ask like they won't reach out. Also, I don't got time to like sit up on a Zoom call and like mentor people. Yeah. >> Uh but if you came ready with like, hey, I read a book on like design designing infrastructure at scale. Um, I read that book early this year and I built I then built contributor info this summer. Um, like that helped navigate myself into like, oh, I should probably set up like job cues. Like in my mind, like, oh, set up a database and like write to it. It's like, oh, well, what happens like two people write at the database at the same time. >> Oh, okay. Well, that makes sense. >> Just clobber the changes. >> Yeah. Just overwrite stuff and like just make it work and like brute force like whatever. Time complexity, algorithm, space, time, red, black tree, whatever. Like who knows when they use that stuff, but it all makes sense when you start queuing jobs and you're like, "Okay, well like what if the same job at the same thing happen at the same time? Which one goes first?" And then how do you like pop stuff off the stack if like you don't need it anymore? >> So I have like a very clever algorithm like if there are PRs that need to be indexed on my project, I will index the freshest data and then I'll go for any data that's missing. Uh so I'm constantly always filling in the gap. So like there's a PR comment that happens later. I'll go fetch that just the comment as a separate job and then I run these jobs literally in parallel. >> Um and it's like literally in my repo I just set up last night a way to paralyze GitHub actions to like kick off these jobs uh and orchestrate all the stuff. And uh I kind of like I I know way too much stuff about GitHub actions and how to leverage it. Uh, but it's extremely useful now because I'm like I would never want to build this myself or never want maintain this, but like >> yeah, >> with like enough like enough grit and determination like you can get some pretty cool technologies out there. >> Yeah. >> So, read books, kids. >> Well said. All right. So, you're not a fan of Will Smith eating spaghetti? >> Man, you were on my Twitter. >> Bringing the heat. >> Yeah, it's like I don't I don't know where this came from, but someone asked in MidJourney or something. No, it wasn't mid Journey. It was like the original Sora or whatever. Like they would generate this Will Smith eating spaghetti and it was just like awful. Like it was just weird. Like a zombie that kind of looked like Will Smith and kind of eating. >> Yeah. >> But it was just like swallowing. And I guess the latest VO3. >> Yeah. >> It could now do a proper Will Smith eating spaghetti. But like how did that become the litmus? Like >> I don't know. >> I think actually a good one is like um uh gymnastics. Uh, so if you if you if you um ask the uh what do you call it? The the LLM to basically generate a video of someone doing like a specific gymnastics experience um or move, it just like starts doing this weird like spiraling like like a quad like a quintuple like backflip or whatever. It's just not realistic. It just starts spinning off the screen. >> Forgets about physics. >> Yeah. Yeah. But the one thing actually if you go further in my my Twitter feed is uh when Sora 2 launched uh someone had a video of they're like they had a um open had a bunch of a string of a bunch of uh videos and the very first one's like hey Sora give me a video of me doing a kick flip and I used to skateboard in high school. It was very clearly a heel flip. So if you don't know about kick flips and heel flips it just spins the other way and I was like that's dumb. Like it's if you're going to worry about strawberries and RS like my tweet was like kick flips are the new R in strawberries. >> Okay. >> Um because they don't know what kick flip is obviously. Uh but I guess the Will Smith eating spaghetti like we saw a problem. Now get the video to do a actual kickflip. Uh and we're good. We make agi. We're there. >> The final boss. Yeah. >> Yeah. Yeah. Kick flips. >> Um yeah. Okay. You also tweeted something about iced coffee. I forget what it was. Do you like iced coffee a lot? >> Uh, probably. I don't know. I don't know the context of that. >> I forget too. I just I don't know. >> Oh, yeah. It was like super hot in San Francisco and like uh like I just get grab an ice coffee and like uh like cool myself off uh and do the rest of my day. But I go to the Blue Bottle in second in in Howard pretty close to there. So, if anybody's in town, >> I'll I'll buy you a coffee. >> I'm going to stalk you there now. >> You buy me a coffee. What am I saying? Yeah. Yeah. >> Yeah. >> Nice. >> Um yeah, that's about all I had. Any other hobbies or interests or things you want to mention? >> Man, hobbies, interest, man. Uh, a couple weeks ago I um I vibe coded a game with my kids. >> I would never build I know what what goes into writing games like even in JavaScript. It's like too much work. Not for me. >> Yeah. >> Uh but what's really cool like if I've coded a game a platformer and then I I went and downloaded a bunch of sprites off like a sprite kit and I'm like cool. Uh and I was actually testing Gemini for this and I was like it actually did a good job on this. like Jim and I take the sprite kit, apply it to my my game engine, and it did a pretty good job. Uh, so I was actually super super proud of that. I do want to actually sit down one weekend and make the game work. Uh, so we built like one, it was like RPG platformer thing. >> Uh, sorry, those are two different things, but basically it's like a platformer, but it was like in the RPG style, we like sort of roam the room >> and like jump and collect wands. So, super simple game because I knew I know >> I know the myth of the mythical man month of like if you have two men it's going to take less time. It's not true. Um but yeah, so what I'm getting at is like the opportunity there is to uh try things that you would not want to try. So >> game development is a new thing. >> Cool. We're going to pause for one second. We're getting some shirts. It looks >> swag. Yeah, I will do it up. >> Uh I'll take an XL, please. >> Oh yeah, I'll do it. >> I will also take an XL. Uh, I will do a large. >> I've been eating a lot of salad, so I used to be XL. >> Yeah, I got to eat more salad. >> Thank you guys. >> Thank you. >> Yeah, thank you. >> So great. >> Yeah. Thanks so much. Yeah, I remember you from the emails. So, yeah. >> Yeah. >> Oh, Jennifer. >> Yes. >> Ah, okay. Yeah. Hey, Nick. >> All right. Wait, which one are you? >> Uh, Nick Taylor. >> Oh, yeah. Brian. I'm the only official one from this at this time. >> Does anybody want a hat? >> I don't need a hat. No. >> Thank you though. >> Thank you guys so much. >> Yeah. Thank you. >> I think we're good. >> Just trying to gonna theoretically do one more after this and then get out of here. So, >> yeah. >> Yeah. Thanks again. >> Yeah. Okay, we're back. So um >> editors editors take all that out >> or don't what you know whatever surprise me. Um okay so what do you want to plug before we end? >> Uh I will plug if you uh are not using AI and you want to use AI in a way that's like self-hosted and like want to leverage in a way that doesn't like sell your data off to a VC. Uh continue. We could run local local models GPTO uh 20B. Actually today we were embargoed. We were working on this like um NGX Spark. So Nvidia shipped like a essentially a Mac Mini sized device. Okay. Okay. >> That runs a GPU inside of it. Uh so it's really cool. >> Super powerful. So you can get like a 320 billion parameter model which >> like 320 that's that's that's pretty large. Um so like a lot of the smaller like sonnetss and stuff like that would like a million u or a billion no trillion. Uh anyway, >> one of those yens. Yeah. Anyway, check out continue, check out contributor.info. I've been doing this open source project. Um, I am 100% intentional not to make a company out of it, but if you are interested in this data, um, come find me. I'd happy to like show you the ropes and how to like navigate open source, but like for growing maintainership, not just sort of finding contribution, but like scaling yourself as maintainer. >> Nice. Nice. Hard problem to solve. So, >> uh, it is >> cool. All right. Thanks for coming on. Thanks for hosting, Nick. Thanks everyone for listening. If you liked it, please subscribe. >> Thanks Josh. >> Leave us some ratings and reviews. We appreciate it. And we will catch you next time., Live at All Things Open W/ Bdougie, Whiskey Web and Whatnot: Web Development, Neat, 01k81wp95nqk9kbbx6krsbv5jk.