Finale & Friends  🔗 **Source:** [Finale & Friends](https://share.snipd.com/episode/8bbdea95-80c4-4a1a-9c4e-e6bec4212d95) ## Highlights - **AI Enabled Rapid Rust Port Of Ladybird** - Ladybird ported its JavaScript engine lib.js from C++ to Rust in ~2 weeks using AI-assisted translation instead of months of manual work. - Andreas Kling leveraged extensive test coverage and tools like Cloud Code and Codex to produce ~25,000 lines of Rust quickly and pragmatically. Transcript: Jerod Santo Yeah. Adam Stacoviak And said, you know what? Swift is actually not what we need. Even after they did a careful evaluation early on, they cracked. They said, you know what? Rust is actually better. And I think the bigger, more unique crack here, which is I kind of expected in this world, was the leveraging of AI to get there, which I thought was quite, I don't know, I guess at some point It's going to be everywhere and it's going to be expected and anticipated that AI is involved in everybody's day-to coding adventures. So it's not surprising, but nice to see they cracked and came back to the good land, which is rust, which is rust. Jerod Santo So a little bit of context here when Andreas Kling and Chris, Chris Wanstroth, who has a difficult name to say quickly, we're on our pod probably last summer, maybe two summers ago. I don't know exactly when it was, but it's when they announced the foundation around Lady Bird and all the stuff they were doing with the money raise and stuff. Or not money raise. What do you call it? Chris donating. I'm just speaking plainly. That whole deal. We asked them about that. And at the time, Andrea said, we're going Swift. Like Lady Bird was written in C++, but they're going to start porting certain portions of it over to Swift. And we released that as a short clip of a longer conversation. And it's become a quite a popular YouTube video where people like to go there and argue about whether or not that was right and why Rust is better and why Swift is better. And why didn't they go with Go? I don't know. There's all kinds of comments on that particular video. And so I was very surprised to see this because Andreas didn't really pull any punches in saying, not that he dislikes Rust, but that he didn't think it was a good fit for a browser because Of how object oriented the DOM is and how browsers work, take his word for it, not mine, and how that Rust doesn't really have the facilities he thinks that he would want in order to do that. And he still hasn't really changed on that. He actually mentions it in his blog post about Lady Bird adopting Rust. He still thinks that, but it's just a pragmatic choice. Rust has the ecosystem. Rust has the momentum. Rust has a lot of other good things about it. Security, of course, for a browser is imperative, and it's so important to the Lady Bird team that it was the pragmatic choice. And that's a curveball. I was just like, didn't see that one coming. I think the way they proved it themselves was less of a curveball when you get into the details. Adam Stacoviak I think when you look at the details, they start with lib.js. That's Ladybird's JavaScript engine. Describes it. Andreas describes it quite well in terms of what it is, but it's got a great it's got extensive test coverage per Andreas, which I think that makes it a good starting point because if you've Got great tests, it's pretty easy for, I would say, an LLM to say, okay, here's a bunch of tests that prove all the things let's just use that as a great way to get to the other side here one Thing he pointed out was there was the result was around 25 000 lines of rust the entire port took around two weeks we had noted that the same work would have taken him about multiple months To do by hand which is kind of obvious to anybody who's been using ai coding or agentic coding or an agent it's pretty obvious that what you can do in a day or two or a week or two with with a Tool like that but 25 000 lines of rust the entire port took around two weeks and the same work would have taken multiple months to do by hand. But the fact they start with a great project from the inside like lib.js with great testing, that is what I think gives you confidence. Like, okay, we've got a great code base with great test coverage. That's a great start to pull in this transition. Jerod Santo Yes. It's is it safe to call andreas a gray beard i know he has a beard i don't think it's necessarily gray the guy's been around and his his title of this post was uh ladybird adopts rust with Help from ai and when i logged this in changelog news for monday i almost pulled that part out because i i searched for the whole thing with ai and i was like well where is the ai i didn't get To that libjs part fast enough he actually specifically names cloud code and codex for the translation i was looking for ai and i thought was he just link baiting is this link bait just ([Time 0:08:06](https://share.snipd.com/snip/d0cc3c88-c15c-4533-a9c7-ed04b6877054)) - **AI Lowers Rust's Mental Overhead** - AI tooling lowers Rust's adoption barrier by handling much of the language's mental overhead and compiler friction. - Jerod Santo notes Rust's complexity (borrow checker, terse syntax) but says AI lets more developers be productive and gain confidence faster. Transcript: Jerod Santo Rust's biggest drawback is its complexity, mental overhead, right? It's hard to learn. Let me just say that. It's very terse. Yes. Adam Stacoviak It requires a lot. The compiler pisses you off. Exactly. Dead code sucks. Yeah, all the things. So you're well aware. Even before you put into prod, you know your issues. That's the thing. You learn how to perform at compile time. Jerod Santo Yes, compile time bugs better than runtime bugs, as all of our static friends will tell us. I'm still a fan of some dynamic languages, but I totally get it. Ruby. Yeah, I'm just going to. Ruby for life, Sorry. I was writing Ruby before I started change logging. I'm going to be writing Ruby after I leave change logging. Ruby for life. That's right. So beautiful. So elegant. So amazing. Yes. Elixir too. So big fans of both of those. But I'm thinking that like, it's such a win for language like Rust to get mass adoption because now we have tools that help us over all of those difficult parts. You know, like eventually, I think if you stick with it, even if you're not hand coding, like you're, you are code reviewing, you are instructing, you are testing, etc. I think you learn it over time. But I think you learn it without any of the showstoppers that certain people of stubbornness make it through. But a lot of us just throw up our hands and say, ([Time 0:13:50](https://share.snipd.com/snip/e44cbfd1-3c88-42c3-a86e-29f0b934c48e)) - **Port Incrementally Using Tests As A Safety Net** - When porting a large codebase, prefer incremental translation using existing tests to minimize disruption and validate correctness. - Adam and Jerod recommend reusing scaffolding and swapping components slowly rather than rewriting everything at once. Transcript: Adam Stacoviak While there was a test coverage, while there was, you know, an existing code base, existing test coverage, why translate from C++ versus new implementation? And I know that he was trying to solve a problem. He was trying to prove the concept probably. But if Rust truly is, and it seems like they're going to keep writing c++ alongside of rust is what he mentions later on the post yeah it's not like a a straight cutoff like this is not the Winner and they're still in like a interrupt kind of scenario where they're doing both boundaries uh but just thinking about it like why not what would have been the process do you think To not translate but rewrite it? Take feature for feature. How could you not translate it versus just say, here's this, examine this code base. We're not translating it. Here's the code, here's the test coverage. Here's all the things it's testing for, which can be intent-based, which is what you hear a lot or what you will hear a lot in this new world working, which is like specs, intent, you know, Document driven. All those are synonymous with, I understand what intent is. I understand what the user experience should be or the DX should be. I understand what I'm trying to get to. Here's where I'm at. My intent is X, which is what he could have done here. What do you think about an intent drivendriven versus, I guess, ported? Yeah, like this way, where it was translated from C++. Why not inspired by versus translated? Jerod Santo That's a really good question. I think that he could answer it better than I could. But what I would surmise is kind of like when we talk about what's going to change here around the change log and the answer is like as little as possible, you know, like I won't be here, But like everything else is going to be pretty much try to stay the same because you don't want, when you have big change, you don't want to have big radical change necessarily. And I think in this case, when you have a big change being introduced into a code base, which is like, here comes a new language and you're proving that out and you're, you're taking steps In that direction. It's like, let's do as little as we can, but still make progress. And so as little change as possible, you're way more likely to get it replaced when you can give the tools the existing code and the tests. And say, let's not rethink the architecture and everything. Let's just, you know, here's our scaffold. Let's pull out this part and let's put in that part and make sure the building still stands. And let's call that a win. And that will help us decide if this is a direction that we want to go. But we can slowly swap that out as we do rethink and rewrite subsections. Just like you're going to make little tweaks here and there to changelog news as you go out making it yours. And so I think that's probably the reason. Probably he just got done way faster that way too. ([Time 0:16:42](https://share.snipd.com/snip/eb53f0d3-ebf2-44e4-9809-c4ebb1a73649)) - **Daily Driver Augie Powers Agent Workflows** - Adam describes using Augment Code's CLI “Augie” daily and praises its context engine for executing well-defined agent flows. - He calls Augie a daily driver that can run tasks unattended when given a clear spec or agent flow. Transcript: Adam Stacoviak Chris, I'm a fan. I use Augie on the daily. It's one of my daily drivers. Now I use Cloud Code. I use Augment Augie. And I also use Amp Code and others. But Augie, I keep going back to it. And here's where I'm at. I feel like not enough of our audience knows about Augment Code. Not enough about Augie, the CLI. It's amazing. I love it. What can you share? Yeah, we often say Augment is the best coding assistant you've never heard of. And that's both frustrating as someone that works there and is very proud of the work we've done, but also inspiring. We want to go and punch above our weight because we aren't anthropic and we aren't open AI. And so the quality of the product itself with our context engine, once you do touch it, people are like just blown away by that. And so like that keeps me going every day. So not to bear the lead here, but this is a paid spot. You are sponsoring this show to get this awareness. Now, at the same time, we're selective and I love to use your tool, but there is in the world. So a lot of developers look at the space and they say, okay, well, how long can this work? How long is this sustainable in the case of Cursor or Windsurf? Or you pick the name and you think discounted tokens, help me shape a lens for our audience. I think it's a lot of awareness, right? Like Cursor got a lot of publicity early on for like fast revenue growth, which well deserved. I think, you know, frankly, some of the media got the, gets the story wrong. And that like, if I gave you a dollar 50 for every dollar you sent me, I'd be the fastest growing startup in the, in the Valley. And so when you're selling discounted tokens, yes, of course you're going to grow very fast. But all that money plus more goes to the model providers. So I think the real story is the story of Anthropic and, you know, being an API provider. I think the market has just moved so fast and there's so many pieces of competition out there that it's just hard to get noticed. So friends, I love Augment Code and I love using Augie and I highly recommend you use it I love using Augie I can hand Augie a well-defined specification a well-defined pep as I call them In my world an agent flow and it executes flawlessly so the cool thing about Augie that I love most really is that context engine and I can hand it a task and it could just churn away on my Well defined plan and just never bother me and accomplish the mission. It is so cool leveraging the latest models, the context engine, and all the fun things behind the scenes in that awesome CLI. So yes, go try it out. Augmentcode.com. Right in the top there is a CLI icon, a terminal icon. Click that, install it, and change your world. ([Time 0:21:20](https://share.snipd.com/snip/a62b5ddf-d9e0-43b0-9fbc-272b26b58663)) - **AI Provider Economics Favor Deep Pockets** - Large model providers face a market tug-of-war where deep pockets (Google) can subsidize offerings while others must find sustainable revenue models. - Jerod and Adam argue Anthropic and OpenAI must monetize APIs or risk being undercut by well-funded incumbents. Transcript: Jerod Santo A lot of blood and water right now between the agents at large you know right there's this whole like clog code subscription thing with open code. I don't know if you've been tracking that, but the way that, yeah. So Anthropic, I think, had an amazing idea with cloud code subscription, which I loved it. Like as soon as Jose added ACP, is that what it's called? Agent client protocol. As soon as I could use my Cloud Code subscription inside Tidewave, I was ready to dive in. Whereas if I had to go out and get a separate source of funding and get an API keys and like spend API tokens. Agent client protocol. Adam Stacoviak That was Zed's thing. Jerod Santo Yeah, Zed started it. And I think that Tidewave is using it in order to use Cloud Code from inside TideWave versus a Claude API token. And I can piggyback that on top of my already existing monthly fixed limit. As a guy who likes to know what he spent every month, I like that. It just makes it an easy adopter. Same thing with these new OpenClaw-style tools. If we can just plug in our codecs, we're already paying whatever, OpenAI, we're already paying Anthropic. Just plug them in, and then it gets rate limited as I have tokens available per month. It's a fixed cost. I feel good about that. Well, Anthropic does not feel good about that. So they've been changing kind of the rules around your Cloud Code subscription and saying that you can't use it for third-party tools. You have to use API tokens. And this has a lot of people not too happy. I'm one of them, but I also totally understand it because, I mean, they got to make some money, right? They got these huge valuations. They got to have revenue, right? Yeah. Adam Stacoviak When you look at the landscape, this is at least where my lens is at. Google has money to burn. And they can operate at zero margin and profit for a long time. Jerod Santo They got a cash cow, which is different. Adam Stacoviak Precisely. And so then you have Anthropic, large investment, large capital raised, massive valuation. You see the big number, but the spend is so astronomical. And then you couple that with opening out the same, right? The same large investments. Yeah. And you got two key players that can't lose, right? That's the game of business. Like don't lose. It's the object, right? Don't lose and play for as long as you can, right? Play forever. ([Time 0:26:08](https://share.snipd.com/snip/1bacae8f-6c2d-41e2-ade6-139e380a2d25)) - **Distillation Attacks Create Cross Border Tensions** - Distillation attacks scrape paid model outputs at scale to train other models, creating legal and ethical tensions across jurisdictions. - Adam summarizes Anthropic's claim of millions of exchanges from fraudulent accounts used to distill Claude's outputs. Transcript: Adam Stacoviak Let's see if they can actually help me out here. So I'm going to paraphrase a little bit of it, but the biggest recent story involving Anthropik, this is me reading directly from Grok's response, so forgive me if there's any corrections Or a lack of correctness the biggest story involving anthropic the company behind cloud code cloud ai models and distillation broke just yesterday which was february 23rd this is February 24th 2026 as we record anthropic publicly accused three major chinese ai companies deep seek moonshot ai and Minimax of running large-scale coordinated distillation attacks On Claude to steal slash improve their own models. Jerod Santo There you go. Adam Stacoviak Now, the world will say our fellow friends out there on Twitter slash X and other places will applaud this. You know, I'm just not, I'm just not sure I want to applaud anything in this tumultuous world where, where, where there's massive downness, I would say in a world where we want sort of Massive upness. Okay. I don't, I'm not sure I want to applaud this. I'm sort of in the middle there because they would say that, well, anthropic did the same thing to other things to get to where they're at. Right know how did their models get trained in the first place right etc and i get that argument but that genie is out of the bottle and and applauding somebody's downfall i suppose because Of somebody else's upfall especially in a coordinated distillation attack i mean would you would you it as an attack? Jerod Santo I suppose so. They are categorized as an attack. I guess it depends on who you are. I think. Adam Stacoviak They said like there was 12 million. I saw this post from 24,000 fake fraudulent accounts, if this is correct. Jerod Santo They were using the slurp down results. Adam Stacoviak Generated over 16 million exchanges, which are conversations slash prompts with Claude. Jerod Santo Yeah, so like, are they, I mean, are they paying? If they're paying, then it's like, I'm sure it's against the terms and they could suspend their accounts. I don't know. It's a tough one. I think that obviously the existing players who have trained their models on the world's information, including proprietary content and the work of artists and the work of photographers And just et cetera, et cetera, and the work of coders. In the case of some of these models trained on publicly available code, but not liberally licensed code, not merely liberally licensed code. And so they're stomping on GPL code, for instance, which we know has happened. I feel like there's recourse for them in the United States, at least. These are all American companies. In the case of Microsoft, OpenAI, Anthropic, and Google, they all can be adjudicated in the court of law. And if there's any justice in our court system, which sometimes you get it, sometimes you don't, I think that will play out. And there has been results based on scanning books, buying books, et cetera. But when it comes to the world stage, I mean, what do you do when it's coming from a different country or different jurisdiction? I don't know. So that's a tough one. ([Time 0:31:08](https://share.snipd.com/snip/e83f742d-5706-4131-b039-0f1c5bf2e7a7)) - **Self Host With Tailscale Funnel For Secure Exposure** - If you want control and fast iteration, try self-hosting critical services and use tools like Tailscale to expose local services securely. - Adam and Jerod recommend Mac Mini or Proxmox HomeLab setups and Tailscale funnels instead of opening ports or relying solely on cloud. Transcript: Adam Stacoviak Let's go self-host. I think, I don't know what you're feeling this year, but I feel like the future this year, this last year was the was the crack in the earth to say on prem is where it's at. I think as we drive more of our own internal tooling, as we start building more of our own SaaS replacements, let's just say. You're going to desire to self-host. And now is the best time ever to get into HomeLab if you're not into that at all. And I think the future is going to be self-hosted slash on-prem. And I'm excited about that. I am so excited about that, as you can probably tell. Jerod Santo Well, I have a little secret I'm happy to reveal, which is that bought a mac mini i bought a mac mini wow yeah was it for our friend open claw well it's well maybe to a certain extent um i've Just been building more stuff you know like it's just living in a world where you can go from idea to something working in minutes. It's just hard not to just build kind of not everything you think of, but just more things. And, you know, I just build stuff here on my laptop and maybe I open source it. Maybe I don't, you know, it runs here. Fine. I end up building stuff that runs on schedules and I'm just like, you know, my laptop closes. This is not a server. This is not a server. This is a work computer, know, that moves places. And I don't have, all I have, the only other computer I have in my entire house, besides laptops for my wife and kids, is a Raspberry Pi that's in the basement that's not plugged in. Adam Stacoviak And you know, I just- right? A Raspberry Pi 3. Yeah. Jerod Santo Still capable. Totally, depending on what you're doing. Well, then I thought, it's just time for some new hardware. And everyone was buying Mac Minis. I kind of got caught up and I told you about my friend that emailed me after, or texted me after the news about OpenClaw. And he was like, just bought a mac mini and i'm just like you know what i'm just gonna i'm gonna buy one i actually bought i actually bought the exact specs that you said you would buy i think On the show that when we talked about it yeah the second tier the top tier it's the top tier oh it's the pro yeah i did not go full ram because i'm not going to run much local i don't know i didn't Want to spend i always go up to ram dude i always want more ram especially when it's big here's the other problem i wanted to play with it and when you go more ram it kicks it back like three Adam Stacoviak Weeks to ship oh yes and i'm like yeah i mean if i'm having tons of fun i don't know maybe i'll buy another one or something and give this one you can bolt on storage you can't bolt on ram yeah And you can bolt on storage at thunderbolt 5 i believe in the mac mini i could be wrong could be thunderbolt 4 so you've got that full bandwidth anyway so i'm going on prem at my point i'm Jerod Santo Going on prem for whatever it is that i have i have nothing to run, but, um, I've been having some fun and I do, I did try out nano claw. I haven't set up open claw, but I do want to know this weird world. All these people are living in and try it for myself. Zero claw actually might be the one I set up first. Cause that's really fast and way less resource heavy. I don't know, but I'm excited to just move some stuff off my laptop over to there and put it on a scheduler. That got me thinking like, you know, Mac's built-in scheduler is LaunchD, which really doesn't have a good interface, you know? And then just in the shower the other day, I was like, I should just build a GUI for it and be sweet. You know, like I could do that in an afternoon with me and my friends. Um, and then it'd be like a cool launch D GUI. Otherwise you can use Cron, but it's a little bit different on Mac and things get weird, you know? So yeah, I'm, uh, I'm all about it, I guess. Adam Stacoviak We'll see. I'd like to see that launch D GUI. Jerod Santo Actually uh that's cool that's cool i'll say you want me to name it i i got stuck on a name i couldn't think of a good name that's why i didn't build it you know you got to have a good name first Adam Stacoviak And then you build the thing but uh i don't know yeah i don't know i can noodle on that a little bit Well, friends, today's podcast is brought to you by our friends over at Squarespace. ([Time 0:40:58](https://share.snipd.com/snip/0f15cc66-8475-4075-bdf8-cfb963d9dc11)) - **Building Turk.run To Escape Slow GitHub Runners** - Adam is building Turk.run, a self-hosted GitHub Actions runner platform to avoid slow hosted runners and painful per-repo token workflows. - He uses Incus containers and Tailscale for networking while dogfooding Turk to build his DNS project. Transcript: Adam Stacoviak I haven't announced it yet, and I'm not sure when I'll actually open source it, but I'm building a self-hosted GitHub runner because I was just tired of how slow GitHub runners are. And it's called Turk. They don't let you run that? Yeah, they do. Yeah, they do. Well, they may change the ball. They may change the plane from that front there. Yeah. But for now, you can build an app and connect to GitHub and build your own. Now, I'm trying to do this. And so the reason why I have so many is because it's in test mode. And it uses Tailscale for the networking to connect to things and so because of that it's got a lot of test machines that are actually not active so these are all dead ones part of my test So truly around 20 devices across my network but in actuality around 54 per the count right now and a lot of that is just cropped from so what's the calendar you're thinking? Jerod Santo Like, you think you ever hit that at your home office slash home lab? Adam Stacoviak No, I think, honestly, I probably hit it within this next year. But at that point, it's propping up either an open source project that has a business attached to it or could have a business attached to it. So I think at that scale, if those were truly sustained, true devices on the tail net uh that's not a home lab anymore i mean i think i don't know maybe actually home prod a home lab with 100 Devices somebody out there's calling father like adam i got 99 man i got it i don't know i guess in the world home lab i'm not really sure yeah i'm not sure i will personally ever have 100 Now if i had new projects i'm building and each one's self-hosted in my home lab and they're self-serving that's a lot of software man i don't know that's tough you want to talk that's Why they made it 100 it's just impossible essentially to hit that number you want to talk more about this runner? Jerod Santo I mean, I'm intrigued. So this is like basically self-host your GitHub runners, right? Adam Stacoviak Yeah. That's the idea. Let's see what I can share now. So there is a, I guess, newer to me, probably less newer to others out there, is a new thing called Incas's new to me again that's why i'm saying new incus incus is i'm on their website uh Linux containers.org/incus incus is a next generation system container application container and virtual machine manager uh some history behind this which i'm'm loosely aware Of is a Canonicals LXD. It was Canonicals LXD. And the, the, the fellow behind this, his name is Stephanie or Stefan. I'm sorry, not Stephanie Stefan, short named just a staff, I believe. And I could be wrong on that front, but I'm trying to get them on the pod in the near future. But this is the underpinning. So when I looked at Incas, and because it's, it supports like any Linux, ZFS, copy on write, you know, it's going to be really easy to build on top of Incas. And so I'm, I'm building Turk right now because I needed to build some, let's just say some stuff on GitHub through runners. And that's super slow. And we've known this because we've been sponsored by Depot and sponsored by Namespace or users of Namespace. And this is well known that GitHub runners are just slow. They're not purpose-built runner platforms to run faster. And so the runner part of it is you have GitHub Actions. Instead of your project, you obviously have.github or something like that. And you have slash workflows. And inside of those workflows are things that will run on GitHub Actions. And so when you push your repo, those live there. And they can run on schedules. You can hit them. They can do your releases. You can tag a release. It'll automate that stuff. And so then that instantiates GitHub Actions, which is essentially a logic platform. An if-then builds Linux, a bunch of crazy stuff. You can run a runner that just checks things, kind of like crawling away too. But then it dumps that off to their own actions runners, which as I mentioned, they're kind of slow, but you can also do self-hosted runners. Okay. And so if you just want to build your own simple self-hosted runner, you can do that today. ([Time 0:54:43](https://share.snipd.com/snip/82e807ee-7a86-4b69-88ce-d0877cab277a)) - **Choose Source Available If You Need Business Protection** - Consider source-available or open-core licensing if you want community visibility while protecting commercial potential. - Adam is torn between permissive open source and source-available models because permissive licensing can enable competitors to absorb his work. Transcript: Adam Stacoviak So Turk is what I built to do that because i was just like i want to self-host runners because get him get him actions runners are just so slow went down that rabbit hole built her and now I'm actually building dns hole with turk and so i'm just so close to like two cool things yeah well that's exciting i'm sure you'll let all us know when these things go out there, right? Jerod Santo I mean, it's ASAP. Adam Stacoviak Why not? So soon. So soon. I actually thought about an image registry as well because another project I have has to have a custom Linux image. And this threw a curb block at me because I was not expecting to have to build a registry for this thing. But a Turk image registry is the next thing. I just started to pin the spec on the Turk image registry. We often laugh, but yeah. Coming soon, Turk.run for now. Well, just know that I'm building it. And if you are interested in, if you believe that self-hosted is the future, which is why I feel so strong about building it. Because I'm like, this can solve our friends' problems. This could solve some enterprises' problems. And I want to build it right to solve my problems, but also think of it in a way that's reusable about the world. So I'm trying to make it an internal project, which is kind of easy in this world, and just say, okay, it'll be internal. I won't do anything with it. I really want to build it in a way that's usable by me. I want to prove it works for me, which was DNS whole being built with it, and then being able to take it to the masses. So that's the next step. ([Time 1:01:56](https://share.snipd.com/snip/75f9f313-1c83-42c6-835a-29bfa8a57e6a)) - **Gate Quality Not Pull Request Volume** - Focus on measurable code quality and automated testing gates rather than manual PR throughput as AI generates many changes. - Adam suggests spinning up many self-hosted runners to battle-test changes and shift review to intent and quality verification. Transcript: Adam Stacoviak What what changes what's the implications of this change is it massive and so in the sense of human review massive change massive prs have been largely frowned upon whereas smaller Changes are easier That's where you get the LGTMs on a massive PR versus a small one, which gets scrutinized to the nth degree, right? Right. Oh, a bike shed. Because it's so much easier to reason about, right? Yeah. And so I think you have this idea of reasoning about the code change, and you care about that. I think we'll always care about some version of code quality, but that's going to change. It's going to be about who. It's the same thing with shift left. It's like, sure, shift left, but now it's just like more on the developer. How much further left can we actually go with the shift left? Same idea there. I think you'll have more to reason about, to think about, and you'll always care about some version of code quality, but the purpose of caring about code quality is is the intent met does It pass tests does it actually work in you know in development environments does it pass the smoke test where we spin it up 50 times and battle test it and that's the point of like where You can do turk type things where you just spin up a ton of runners because they're more free yeah in a self-hosted world. So you wouldn't spin up 50 runners on GitHub to test because one, potentially super expensive. Two, my gosh, super slow. But in the world of Turk, where you can self-host and infinitely test on easily clonable VMs or system containers, that's the world you live in. So you can ensure code quality at a much higher degree. And I think AI will continue to evolve this and will care about what has been code review, I think will translate into some version of code quality. And I'm actually wondering if the next big thing on this front isn't some version of a code quality gateway. You see this happening in AI where you have AI gateways, where you want to gateway the AI. This happened in networking, right? You have a network gateway or something like that where you can log into something. Jerod Santo Open Router is a good example that you're talking about, right? Adam Stacoviak Exactly. Something that lets you access more of the network is a gateway. I think I have more of that in the code quality and what has traditionally been code review is going to be more in the real time. I want to know just in time, not before I push and hope and pray, but I just built this feature. Does this meet my intent? How does it work with others? I want to see it on my machine, not push to some place and hope and pray. I think things will change on that front there. ([Time 1:14:02](https://share.snipd.com/snip/94b36236-3e2f-47fa-8b96-d1c6e6e7f7a6)) - **AI Compresses The SDLC Into Real Time Loops** - Boris Tane's claim that the SDLC is dying highlights how AI compresses previously serial development stages into real-time loops. - Jerod and Adam see code review shifting toward just-in-time quality gates and intent checks rather than ritualized human queues. Transcript: Adam Stacoviak Two more things on our list. They're both kind of deep, and I'm not sure how deep you went with them because I shared this list with you moments before the show. I was just going to ride your coattails, which is, you know, dangerous, but I do like to live a little bit dangerous. It could be dangerous. Okay. So I'm going to, I'm going to loosely mention Boris Tain. So Boris Tain is going to come on the pod soon. The invite has been accepted. The plan has not been made. So there's not a date on the calendar yet. But one thing he said recently was the software development life cycle is dead. You often hear the term, Jared, if you haven't, this is new to me. I learned it recently, SDLC. I'm like, what is SDLC? Oh, software development lifecycle. So only if you sell SDLC type products do you know SDLC acronyms, right? Right. Because you don't want to keep saying software development lifecycle. Jerod Santo It's a lot. Adam Stacoviak Right. Jerod Santo It's a lot. Adam Stacoviak So i get it i get the acronym but for a while i was like okay what does sdlc mean again okay it kind of makes sense but he proposes i should say he purports that the sdlc as you've learned it Is a relic he shares let's see we can throw some video up on this but he shares a classic software development like cycle where we're taught requirements system design implementation Testing code review hello deployment monitoring and recycle back to requirements again like this has been a known loop and he goes on to suggest how the uh every stage is collapsing How ai is really collapsing a lot of this and it's not because ai is eating it it's because more of what happened in the sdlc is happening more once and in real time versus in different disparate Places and i think that's why i really care about just in time code review as well and it's not really code review it's just more like dude is my code okay? You know? I actually want a server to call that dudeismycodeokay.com. And then that'd be kind of cool. I want that, right? Because I don't really want code review. But it kind of goes on through this. And I think that kind of leans into the fact that we just talked about code review and code quality. Yeah. That's going to be the thing. It's going to be more this is getting compressed more this is getting uh replaced and compartmentalized in a way that it's just like it's just collapsing a lot of this is collapsing and I i kind of mourn in a way the ([Time 1:16:43](https://share.snipd.com/snip/88203360-911f-427f-b2da-9fe77cc6ec67)) - **Wes McKinney Found Coding More Fun With Better Tools** - Wes McKinney said coding is more fun now because tools free him to think what to build instead of wrestling with tooling. - Jerod echoes this, enjoying rapid prototyping and throwing away experiments that don't stick. Transcript: Jerod Santo I found that to be absolutely the case with me, just to give my amen to what you just said and to back it up. It's just fun to work at this level and just touch all the things, try all the things, build something, see if it sticks. ([Time 1:34:54](https://share.snipd.com/snip/e2618dcf-8193-48e5-b661-435c3dee3191)) - **Doom Futures Need Counterarguments** - Fictional scenarios like the 2028 Global Intelligence Crisis present bearish futures of automation-driven economic disruption and provoke reading both optimistic and pessimistic takes. - Adam recommends pairing the crisis narrative with Anish Acharya's rebuttal to avoid getting unduly alarmed. Transcript: Adam Stacoviak And then I want you to also check out Anish. I'm not sure how you say his last name. Acharya, maybe. He is from A16Z and Dreesen Horwitz. And there was a, you may have seen this, Jared, and tell me if you did, but there was a post that circulated called the 2028 Global Intelligence Crisis. Does this ring a bell to you at all? Jerod Santo No. Tell me about it. Adam Stacoviak Well, it is a thought exercise in financial history from the future. And so it actually writes itself. The very first, the dated line is February 22nd, 2026. That's scratched out. And it june 30th 2028 okay it's largely you know bearish not bullish and negative but it's it's they say it's grounded on some research and it's essentially the consequences of abundant Intelligence and it paints this picture of a of a just a downside that's coming when it comes to what automation does and i'm just not sure i'm just not sure yeah uh i'm it's all i'm gonna Say because it really is that deep it's a whole podcast worthy uh topic but one thing that a niche shared was a different way that this could go down i'm only got both because i don't want Anybody out there reading this thing uh and getting all negative and sad and going into a hole when we say go touch all the things. And then touch all the things turns into all this nasty stuff that this seemingly well-researched article from the future purports. So I would read Anisha's prose on X, which we'll link to in tandem with this and try your best to come out not crying like a little baby. Because it is pretty wild. I'll read the first line. Okay. The unemployment rate printed 10.2% this morning, a 0.3% upside surprise. The market sold off 2% of the number on the number, bringing the cumulative down draw in the S&P to 38% from its October 2026 highs. ([Time 1:35:44](https://share.snipd.com/snip/f2c160d2-b8dd-4f08-932c-433e34f0dd4e))