[
  {
    "startMs": 1309,
    "text": "[music]"
  },
  {
    "startMs": 12719,
    "text": "Hello, my name is Jonathan Kelly and"
  },
  {
    "startMs": 15759,
    "text": "today we're going to talk about what it"
  },
  {
    "startMs": 17199,
    "text": "means to build ambitious software in the"
  },
  {
    "startMs": 19760,
    "text": "age of AI."
  },
  {
    "startMs": 22640,
    "text": "5 years ago, I made the first commit"
  },
  {
    "startMs": 24880,
    "text": "ever to a project called Diosis. I used"
  },
  {
    "startMs": 28000,
    "text": "the last summer I had as an"
  },
  {
    "startMs": 29359,
    "text": "undergraduate and instead of getting an"
  },
  {
    "startMs": 31279,
    "text": "internship at Google or doing research"
  },
  {
    "startMs": 33120,
    "text": "in AI like many of my friends at the"
  },
  {
    "startMs": 35280,
    "text": "time, I spent it exploring an idea I had"
  },
  {
    "startMs": 38000,
    "text": "for a crossplatform app framework"
  },
  {
    "startMs": 40960,
    "text": "written in the Rust programming"
  },
  {
    "startMs": 42640,
    "text": "language."
  },
  {
    "startMs": 44160,
    "text": "In 2021, Rust was still pretty niche,"
  },
  {
    "startMs": 47840,
    "text": "but the ecosystem was growing, the"
  },
  {
    "startMs": 49680,
    "text": "tooling was improving, and the pitch of"
  },
  {
    "startMs": 51680,
    "text": "native performance, a solid type system,"
  },
  {
    "startMs": 54320,
    "text": "and simple cross compilation really sold"
  },
  {
    "startMs": 56480,
    "text": "me. It's extremely nerdy."
  },
  {
    "startMs": 60239,
    "text": "The idea for Diosis was straightforward."
  },
  {
    "startMs": 63120,
    "text": "What if we had an crossplatform app"
  },
  {
    "startMs": 65920,
    "text": "framework? Instead of waiting through"
  },
  {
    "startMs": 68080,
    "text": "dozens of tool chains, programming"
  },
  {
    "startMs": 69920,
    "text": "languages, and idees, what if we simply"
  },
  {
    "startMs": 72960,
    "text": "wrote our all of our apps in Rust using"
  },
  {
    "startMs": 76240,
    "text": "HTML and CSS as the markup language?"
  },
  {
    "startMs": 79360,
    "text": "This was back in the day 2021 where"
  },
  {
    "startMs": 81840,
    "text": "React Native was janky, Flutter was too"
  },
  {
    "startMs": 84240,
    "text": "slow, and neither performed well with"
  },
  {
    "startMs": 86159,
    "text": "native APIs. On the flip, with Rust, we"
  },
  {
    "startMs": 90240,
    "text": "could build native abstractly with no"
  },
  {
    "startMs": 92320,
    "text": "VM, no IPC, no JavaScript. And if we"
  },
  {
    "startMs": 96640,
    "text": "used a little bit of HTML and CSS for"
  },
  {
    "startMs": 99280,
    "text": "the UI and take some inspiration from"
  },
  {
    "startMs": 101520,
    "text": "React for the reactivity, we could reuse"
  },
  {
    "startMs": 104000,
    "text": "vast amounts of web components and web"
  },
  {
    "startMs": 106000,
    "text": "tooling. The goal was an extremely"
  },
  {
    "startMs": 108240,
    "text": "powerful app framework that was still"
  },
  {
    "startMs": 110000,
    "text": "quite familiar to the average developer."
  },
  {
    "startMs": 113600,
    "text": "Sounds easy, right? Well, as they say,"
  },
  {
    "startMs": 117040,
    "text": "we choose to build an app framework from"
  },
  {
    "startMs": 118880,
    "text": "scratch, not because it's easy, but"
  },
  {
    "startMs": 120719,
    "text": "because we thought it would be easy."
  },
  {
    "startMs": 123360,
    "text": "In reality, trying to challenge React"
  },
  {
    "startMs": 124880,
    "text": "Native and Flutter is extremely"
  },
  {
    "startMs": 126240,
    "text": "ambitious. In 2021, there were very few"
  },
  {
    "startMs": 129440,
    "text": "off-the-shelf components you could use"
  },
  {
    "startMs": 131200,
    "text": "to build Dioxis. Everything from"
  },
  {
    "startMs": 133440,
    "text": "reactivity to font rendering to hot"
  },
  {
    "startMs": 136000,
    "text": "reloading and application bundling had"
  },
  {
    "startMs": 138720,
    "text": "to be built from scratch. There's"
  },
  {
    "startMs": 140640,
    "text": "nothing we could use. For us, tasks like"
  },
  {
    "startMs": 143920,
    "text": "building a web browser were just"
  },
  {
    "startMs": 145920,
    "text": "necessary steps along the way."
  },
  {
    "startMs": 149920,
    "text": "Now today in 2026, Diosis has achieved"
  },
  {
    "startMs": 153280,
    "text": "and far surpassed its original mission."
  },
  {
    "startMs": 156239,
    "text": "We support all the features we"
  },
  {
    "startMs": 157760,
    "text": "originally set out to build from"
  },
  {
    "startMs": 159519,
    "text": "crossplatform support to native"
  },
  {
    "startMs": 161440,
    "text": "rendering to Rust hot reload and bundle"
  },
  {
    "startMs": 164160,
    "text": "splitting. We've basically reinvented"
  },
  {
    "startMs": 167040,
    "text": "and improved the entire app development"
  },
  {
    "startMs": 169519,
    "text": "stack."
  },
  {
    "startMs": 170879,
    "text": "Users can ship a powerful full stack web"
  },
  {
    "startMs": 173599,
    "text": "application in the exact same codebase"
  },
  {
    "startMs": 176239,
    "text": "sharing components as their iOS and"
  },
  {
    "startMs": 178959,
    "text": "Android apps."
  },
  {
    "startMs": 180959,
    "text": "The Daxis project now has nearly 37,000"
  },
  {
    "startMs": 184239,
    "text": "stars on GitHub with millions of"
  },
  {
    "startMs": 185920,
    "text": "downloads. Apps built in Dioxis are"
  },
  {
    "startMs": 188480,
    "text": "rolled across the globe with a"
  },
  {
    "startMs": 191120,
    "text": "cumulative estimate of over 200 million"
  },
  {
    "startMs": 192879,
    "text": "end users. Users have built things like"
  },
  {
    "startMs": 195519,
    "text": "AI assistance, software for voting, data"
  },
  {
    "startMs": 199040,
    "text": "science tools, and even collision"
  },
  {
    "startMs": 201200,
    "text": "avoidance system for satellites in"
  },
  {
    "startMs": 203440,
    "text": "space."
  },
  {
    "startMs": 205760,
    "text": "We've put a ton of effort into making"
  },
  {
    "startMs": 207599,
    "text": "Dioxis as userfriendly as possible."
  },
  {
    "startMs": 210319,
    "text": "Fewer files, build tooling, hot"
  },
  {
    "startMs": 212959,
    "text": "reloading, asset optimization,"
  },
  {
    "startMs": 215040,
    "text": "everything you need to easily ship"
  },
  {
    "startMs": 216400,
    "text": "across all platforms."
  },
  {
    "startMs": 218480,
    "text": "Because Dioxis apps are written in Rust,"
  },
  {
    "startMs": 220400,
    "text": "they are structurally very simple. You"
  },
  {
    "startMs": 223599,
    "text": "rarely need to drop into platform"
  },
  {
    "startMs": 225360,
    "text": "specific code because all rest projects"
  },
  {
    "startMs": 228000,
    "text": "are alike. It's very easy for developers"
  },
  {
    "startMs": 230080,
    "text": "to dive into a new project. You can"
  },
  {
    "startMs": 232080,
    "text": "completely skip annoying build system"
  },
  {
    "startMs": 233920,
    "text": "setup. All you need is a main.rs to get"
  },
  {
    "startMs": 237040,
    "text": "started."
  },
  {
    "startMs": 239840,
    "text": "One of the most ambitious goals we had"
  },
  {
    "startMs": 241439,
    "text": "for ship lightweight but fully featured"
  },
  {
    "startMs": 245040,
    "text": "HTML and CSS rendering engine called"
  },
  {
    "startMs": 247040,
    "text": "Blitz. We extracted the browser grade"
  },
  {
    "startMs": 249760,
    "text": "CSS engine out of Firefox, built our own"
  },
  {
    "startMs": 252560,
    "text": "HTML DOM, and developed a hybrid GPU"
  },
  {
    "startMs": 255840,
    "text": "rendering pipeline."
  },
  {
    "startMs": 257759,
    "text": "Compared to Electron apps, which are RAM"
  },
  {
    "startMs": 259840,
    "text": "and storage hogs, Blitz apps are"
  },
  {
    "startMs": 261840,
    "text": "lightweight, coming in at less than 5"
  },
  {
    "startMs": 263680,
    "text": "megabytes bundle sizes and consume less"
  },
  {
    "startMs": 265919,
    "text": "than 50 megabytes of RAM at runtime. And"
  },
  {
    "startMs": 268639,
    "text": "they're pretty cool. You can write your"
  },
  {
    "startMs": 269840,
    "text": "own custom components, spinning cubes,"
  },
  {
    "startMs": 271840,
    "text": "you can customize the browser however"
  },
  {
    "startMs": 272960,
    "text": "you want. It's a very cool project."
  },
  {
    "startMs": 275440,
    "text": "We also worked on a tool called"
  },
  {
    "startMs": 276720,
    "text": "Subsecond uh which is our generic hot"
  },
  {
    "startMs": 279360,
    "text": "reload engine for Rust, C and C++."
  },
  {
    "startMs": 282560,
    "text": "Subsecond watches your code for edits,"
  },
  {
    "startMs": 284639,
    "text": "recompiles parts of the code that"
  },
  {
    "startMs": 286080,
    "text": "changed and patches the running app in"
  },
  {
    "startMs": 288479,
    "text": "place all in 100 100 milliseconds. This"
  },
  {
    "startMs": 291919,
    "text": "was an incredibly difficult technical"
  },
  {
    "startMs": 293520,
    "text": "challenge and is the only hot reload"
  },
  {
    "startMs": 295440,
    "text": "engine for native compiled code to have"
  },
  {
    "startMs": 298000,
    "text": "such wide language and runtime support."
  },
  {
    "startMs": 300240,
    "text": "It works on every major system and even"
  },
  {
    "startMs": 302479,
    "text": "the web is compiled into web assembly."
  },
  {
    "startMs": 305120,
    "text": "No one has done this before"
  },
  {
    "startMs": 309280,
    "text": "because these projects we've worked on"
  },
  {
    "startMs": 311120,
    "text": "along the way over the past 5 years are"
  },
  {
    "startMs": 313199,
    "text": "incredibly ambitious and are the result"
  },
  {
    "startMs": 315199,
    "text": "of a tiny but"
  },
  {
    "startMs": 320560,
    "text": "over every line of code by with our own"
  },
  {
    "startMs": 323600,
    "text": "two eyes and maintain a frequent but"
  },
  {
    "startMs": 325440,
    "text": "ambitious release cadence."
  },
  {
    "startMs": 328080,
    "text": "The most amazing thing, every line of"
  },
  {
    "startMs": 330320,
    "text": "code in Dioxis until very recently has"
  },
  {
    "startMs": 333440,
    "text": "been painstakingly written by hand."
  },
  {
    "startMs": 337280,
    "text": "Why do I say recently? Well, if you"
  },
  {
    "startMs": 340000,
    "text": "aren't aware, software engineering and"
  },
  {
    "startMs": 342800,
    "text": "development has taken a massive turn in"
  },
  {
    "startMs": 344800,
    "text": "the past 6 months. AI coding agents got"
  },
  {
    "startMs": 348160,
    "text": "really, really good. And specifically,"
  },
  {
    "startMs": 350240,
    "text": "they got really good at Rust. Our team,"
  },
  {
    "startMs": 353759,
    "text": "a bunch of cracked rest engineers, has"
  },
  {
    "startMs": 356160,
    "text": "been quite skeptical of AI for a long"
  },
  {
    "startMs": 358240,
    "text": "time. We had not felt the AGI, so to"
  },
  {
    "startMs": 361600,
    "text": "speak. And we definitely weren't using"
  },
  {
    "startMs": 363520,
    "text": "AI in our day-to-day work. We thought"
  },
  {
    "startMs": 365840,
    "text": "the two things were incompatible,"
  },
  {
    "startMs": 367360,
    "text": "shipping high quality code"
  },
  {
    "startMs": 370240,
    "text": "tools. Seeing them get really good at"
  },
  {
    "startMs": 372800,
    "text": "Rust was a huge surprise to us. So, we"
  },
  {
    "startMs": 375199,
    "text": "were finally excited."
  },
  {
    "startMs": 378479,
    "text": "With this newfound excitement, we"
  },
  {
    "startMs": 380160,
    "text": "started building our team. maxed out our"
  },
  {
    "startMs": 382400,
    "text": "cloud code subscriptions, turned out"
  },
  {
    "startMs": 384160,
    "text": "tens of thousands of lines of Rust, and"
  },
  {
    "startMs": 386000,
    "text": "built all sorts of features we had long"
  },
  {
    "startMs": 387520,
    "text": "wished to have. Unfortunately, very"
  },
  {
    "startMs": 390560,
    "text": "little of the code cleared our quality"
  },
  {
    "startMs": 392639,
    "text": "bar of should we merge this in."
  },
  {
    "startMs": 395440,
    "text": "Thousands of lines of new features, bug"
  },
  {
    "startMs": 397759,
    "text": "fixes, and integrations we had wanted"
  },
  {
    "startMs": 400319,
    "text": "for years sat there and draft and"
  },
  {
    "startMs": 402880,
    "text": "continue to sit there and draft. We"
  },
  {
    "startMs": 405600,
    "text": "definitely did not know how to properly"
  },
  {
    "startMs": 407039,
    "text": "wield these tools and it was way too"
  },
  {
    "startMs": 409919,
    "text": "easy to become what we call a slop"
  },
  {
    "startMs": 412880,
    "text": "cannon."
  },
  {
    "startMs": 414639,
    "text": "Um, so we reflected a bit and studied"
  },
  {
    "startMs": 417360,
    "text": "what worked and what didn't."
  },
  {
    "startMs": 427759,
    "text": "to read, easy to write, good tools, good"
  },
  {
    "startMs": 430319,
    "text": "error messages. The coding agents"
  },
  {
    "startMs": 432560,
    "text": "generally don't care about this. We"
  },
  {
    "startMs": 434639,
    "text": "tried to make Rust easel"
  },
  {
    "startMs": 440080,
    "text": "with Dioxis still"
  },
  {
    "startMs": 442639,
    "text": "because Rust is harder to write. The"
  },
  {
    "startMs": 444880,
    "text": "coding agents deal with the development"
  },
  {
    "startMs": 446240,
    "text": "burden for you. They handle the edge"
  },
  {
    "startMs": 448240,
    "text": "cases and they fight the borrow checker,"
  },
  {
    "startMs": 450479,
    "text": "saving you from the cognitive burden of"
  },
  {
    "startMs": 452800,
    "text": "writing Rust apps. The learning curve"
  },
  {
    "startMs": 455520,
    "text": "which we fought to reduce is now a"
  },
  {
    "startMs": 458080,
    "text": "feature."
  },
  {
    "startMs": 460240,
    "text": "So throughout the process of adopting"
  },
  {
    "startMs": 463280,
    "text": "the coding tools to work on Dioxis, we"
  },
  {
    "startMs": 465520,
    "text": "learned a wide array of lessons."
  },
  {
    "startMs": 468560,
    "text": "Many of the thing the coding agents do"
  },
  {
    "startMs": 470560,
    "text": "really well today and many things they"
  },
  {
    "startMs": 472800,
    "text": "just aren't there yet. So the next"
  },
  {
    "startMs": 475280,
    "text": "couple slides I want to talk about some"
  },
  {
    "startMs": 476720,
    "text": "of the things we learned and what it"
  },
  {
    "startMs": 478160,
    "text": "means to to build ambitious software"
  },
  {
    "startMs": 480960,
    "text": "projects in the age of agent coding."
  },
  {
    "startMs": 485039,
    "text": "Um, it's important to talk about first"
  },
  {
    "startMs": 487919,
    "text": "what it means to build an ambitious"
  },
  {
    "startMs": 489680,
    "text": "software project. Um, there's many"
  },
  {
    "startMs": 492879,
    "text": "different types of software out there."
  },
  {
    "startMs": 494720,
    "text": "Depends on what you ship every day. Um,"
  },
  {
    "startMs": 497280,
    "text": "you might be doing research and the"
  },
  {
    "startMs": 500080,
    "text": "quality of your code isn't the most"
  },
  {
    "startMs": 501919,
    "text": "important thing. You might be doing"
  },
  {
    "startMs": 502960,
    "text": "prototyping code and and iterating fast,"
  },
  {
    "startMs": 505599,
    "text": "moving quickly is important. You might"
  },
  {
    "startMs": 507520,
    "text": "be building applications which people"
  },
  {
    "startMs": 509199,
    "text": "don't see the code internally. They just"
  },
  {
    "startMs": 511440,
    "text": "see what it looks like on the outside."
  },
  {
    "startMs": 514000,
    "text": "But for us and for Diosis,"
  },
  {
    "startMs": 517440,
    "text": "we care about a few different things."
  },
  {
    "startMs": 520560,
    "text": "Primarily, of course, we care that our"
  },
  {
    "startMs": 522320,
    "text": "code works all the time and that if it"
  },
  {
    "startMs": 524800,
    "text": "breaks, we can easily fix it. I think"
  },
  {
    "startMs": 527440,
    "text": "this is something people don't think"
  },
  {
    "startMs": 528720,
    "text": "about enough these days that you need to"
  },
  {
    "startMs": 530959,
    "text": "continue to build easily maintainable"
  },
  {
    "startMs": 532800,
    "text": "code and like the velocity that you ship"
  },
  {
    "startMs": 536160,
    "text": "lays down on this substrate that you've"
  },
  {
    "startMs": 538640,
    "text": "built and if the substrate isn't good,"
  },
  {
    "startMs": 540560,
    "text": "nothing you build on top is going to be"
  },
  {
    "startMs": 542080,
    "text": "good."
  },
  {
    "startMs": 543839,
    "text": "Secondarily, we care about shipping new"
  },
  {
    "startMs": 546080,
    "text": "features."
  },
  {
    "startMs": 547680,
    "text": "Our road map is really long. It extends"
  },
  {
    "startMs": 550399,
    "text": "into the far future. There's dozens of"
  },
  {
    "startMs": 553200,
    "text": "features we still have yet to build for"
  },
  {
    "startMs": 555360,
    "text": "Dioxis. And we want to ship these"
  },
  {
    "startMs": 557519,
    "text": "quickly to keep up with the times, but"
  },
  {
    "startMs": 560640,
    "text": "we also want to maintain quality. When"
  },
  {
    "startMs": 563120,
    "text": "building a large ambitious project like"
  },
  {
    "startMs": 565200,
    "text": "Deiosis, there's a constant tension of"
  },
  {
    "startMs": 567680,
    "text": "shipping fast, adding new features, and"
  },
  {
    "startMs": 570480,
    "text": "then also making sure you don't break"
  },
  {
    "startMs": 571920,
    "text": "things and that in a patch release,"
  },
  {
    "startMs": 573519,
    "text": "you're not like breaking APIs that"
  },
  {
    "startMs": 576160,
    "text": "millions of people rely on. For a"
  },
  {
    "startMs": 578959,
    "text": "project that people build their"
  },
  {
    "startMs": 580160,
    "text": "businesses on, there's also a high bar"
  },
  {
    "startMs": 582000,
    "text": "for releases. We need to maintain uh"
  },
  {
    "startMs": 584800,
    "text": "high quality of our documentation, of"
  },
  {
    "startMs": 587440,
    "text": "our examples, of our tests, of our"
  },
  {
    "startMs": 589279,
    "text": "benchmarks. If it if anything is out of"
  },
  {
    "startMs": 591760,
    "text": "place, people figure it out pretty"
  },
  {
    "startMs": 593760,
    "text": "quickly."
  },
  {
    "startMs": 596640,
    "text": "So, we really do like coding agents as a"
  },
  {
    "startMs": 601120,
    "text": "a form of an excellent assistant for"
  },
  {
    "startMs": 603120,
    "text": "very hard technical problems. Coding"
  },
  {
    "startMs": 605519,
    "text": "agents bring a level of patience and"
  },
  {
    "startMs": 607760,
    "text": "massive knowledge that is very hard to"
  },
  {
    "startMs": 610320,
    "text": "muster as an individual working on a"
  },
  {
    "startMs": 612240,
    "text": "very large software project."
  },
  {
    "startMs": 614560,
    "text": "Many problems in Dioxis are knowledge"
  },
  {
    "startMs": 616959,
    "text": "problems. Our team can't feasibly know"
  },
  {
    "startMs": 619839,
    "text": "every detail about every build system,"
  },
  {
    "startMs": 621839,
    "text": "every runtime, every operating system,"
  },
  {
    "startMs": 624079,
    "text": "every programming language, every API,"
  },
  {
    "startMs": 626880,
    "text": "every quirk. Fortunately, this is"
  },
  {
    "startMs": 629600,
    "text": "exactly where the coding agents excel."
  },
  {
    "startMs": 632160,
    "text": "They can quickly quickly sift through"
  },
  {
    "startMs": 634079,
    "text": "thousands of pages of documentation,"
  },
  {
    "startMs": 636560,
    "text": "read all the bespoke APIs, dig into"
  },
  {
    "startMs": 639920,
    "text": "binaries, reverse engineer APIs. They"
  },
  {
    "startMs": 642480,
    "text": "have so much more patience than an"
  },
  {
    "startMs": 644480,
    "text": "individual developer does."
  },
  {
    "startMs": 646959,
    "text": "We were able to implement things like"
  },
  {
    "startMs": 648959,
    "text": "cotlin and swift plugins for dioxysis"
  },
  {
    "startMs": 651839,
    "text": "deeply integrated into our build system"
  },
  {
    "startMs": 654079,
    "text": "which is a really hard feature. If you"
  },
  {
    "startMs": 656320,
    "text": "know react native turbo modules these"
  },
  {
    "startMs": 658959,
    "text": "things took many years of development to"
  },
  {
    "startMs": 660959,
    "text": "get right by people writing them by"
  },
  {
    "startMs": 662959,
    "text": "hand. We were able to ship this in like"
  },
  {
    "startMs": 665440,
    "text": "two to three weeks with coding agents"
  },
  {
    "startMs": 667120,
    "text": "and we probably could have gone faster."
  },
  {
    "startMs": 668399,
    "text": "I think implementation was done in like"
  },
  {
    "startMs": 670000,
    "text": "the first day and we spent two weeks"
  },
  {
    "startMs": 672000,
    "text": "building test cases and testing on real"
  },
  {
    "startMs": 673839,
    "text": "devices."
  },
  {
    "startMs": 675519,
    "text": "And in Blitz, the thing on the right,"
  },
  {
    "startMs": 677760,
    "text": "our custom web engine, web agents have"
  },
  {
    "startMs": 680800,
    "text": "accelerated debugging hard CSS, styling,"
  },
  {
    "startMs": 683519,
    "text": "and layout issues for us. The agents"
  },
  {
    "startMs": 685839,
    "text": "know the CSS spec exceptionally well."
  },
  {
    "startMs": 688320,
    "text": "You might be writing a line of code"
  },
  {
    "startMs": 689600,
    "text": "that's trying to resolve some sort of"
  },
  {
    "startMs": 691760,
    "text": "painting or layout issue. And the agents"
  },
  {
    "startMs": 693760,
    "text": "can instantly recall exactly how Google"
  },
  {
    "startMs": 695680,
    "text": "Chrome and Safari do it. Can tell you"
  },
  {
    "startMs": 697920,
    "text": "the right way of handling it for your"
  },
  {
    "startMs": 699600,
    "text": "problem. and you don't have to go open"
  },
  {
    "startMs": 701600,
    "text": "the the WebKit source code that's nested"
  },
  {
    "startMs": 704480,
    "text": "deep somewhere in Apple's Git"
  },
  {
    "startMs": 707120,
    "text": "repositories."
  },
  {
    "startMs": 709120,
    "text": "We're able to invest time in doing"
  },
  {
    "startMs": 711920,
    "text": "things the right way, not the hacky way,"
  },
  {
    "startMs": 714800,
    "text": "which interestingly is a turn and"
  },
  {
    "startMs": 717120,
    "text": "compared to how we used to do it. We"
  },
  {
    "startMs": 719279,
    "text": "would always gauge a project based on"
  },
  {
    "startMs": 720640,
    "text": "its complexity and tend to take"
  },
  {
    "startMs": 722399,
    "text": "shortcuts as humans to ship things"
  },
  {
    "startMs": 725279,
    "text": "faster but not at a high quality bar. So"
  },
  {
    "startMs": 728240,
    "text": "coding agents give us the ability to"
  },
  {
    "startMs": 730480,
    "text": "maintain quality and do things the right"
  },
  {
    "startMs": 732240,
    "text": "way which is very interesting."
  },
  {
    "startMs": 735279,
    "text": "Um a less sexy application of coding"
  },
  {
    "startMs": 738399,
    "text": "agents for ambitious projects is"
  },
  {
    "startMs": 740639,
    "text": "actually doing the extremely mundane"
  },
  {
    "startMs": 742320,
    "text": "tasks. Uh our team is very small. We"
  },
  {
    "startMs": 744959,
    "text": "have three core engineers working on"
  },
  {
    "startMs": 746399,
    "text": "Diosis. uh any time that we spend like"
  },
  {
    "startMs": 749839,
    "text": "verifying the tarball extracts into the"
  },
  {
    "startMs": 752000,
    "text": "right directory structure is like time"
  },
  {
    "startMs": 754000,
    "text": "wasted from us thinking about the"
  },
  {
    "startMs": 755279,
    "text": "architecture and the the hard problems"
  },
  {
    "startMs": 757120,
    "text": "of our software. Um Dioxis is a large"
  },
  {
    "startMs": 760240,
    "text": "project and it's been a challenge to"
  },
  {
    "startMs": 762320,
    "text": "maintain a high quality bar across the"
  },
  {
    "startMs": 764399,
    "text": "entire codebase across every release. In"
  },
  {
    "startMs": 766959,
    "text": "one release we might add an extension"
  },
  {
    "startMs": 768800,
    "text": "for a new editor like zed. We might not"
  },
  {
    "startMs": 771440,
    "text": "be able to test that editor every time"
  },
  {
    "startMs": 773360,
    "text": "we do a patch release. And it might be"
  },
  {
    "startMs": 774720,
    "text": "easy to break that. Applying agents to"
  },
  {
    "startMs": 777360,
    "text": "the problem actually lets us automate"
  },
  {
    "startMs": 778800,
    "text": "many of these like hard tedious tasks"
  },
  {
    "startMs": 780480,
    "text": "that would have taken like countless"
  },
  {
    "startMs": 782480,
    "text": "hours before."
  },
  {
    "startMs": 784320,
    "text": "And then for us like the code is the"
  },
  {
    "startMs": 786959,
    "text": "product. People download the code, they"
  },
  {
    "startMs": 788959,
    "text": "build on the code, users interact with"
  },
  {
    "startMs": 791040,
    "text": "our APIs, they read our docs, and they"
  },
  {
    "startMs": 792800,
    "text": "build on our architecture. So any"
  },
  {
    "startMs": 794880,
    "text": "laziness in the the quality of the code,"
  },
  {
    "startMs": 797519,
    "text": "the SDKs that we ship to users"
  },
  {
    "startMs": 800399,
    "text": "translates directly into a worse"
  },
  {
    "startMs": 802160,
    "text": "developer experience and people either"
  },
  {
    "startMs": 804240,
    "text": "getting upset, their businesses being"
  },
  {
    "startMs": 806160,
    "text": "stalled, or them turning off the"
  },
  {
    "startMs": 807839,
    "text": "product. So coding agents have been"
  },
  {
    "startMs": 810480,
    "text": "excellent at maintaining uh tasks like"
  },
  {
    "startMs": 813600,
    "text": "verifying release checklists,"
  },
  {
    "startMs": 815200,
    "text": "backporting bug fixes onto stable"
  },
  {
    "startMs": 816959,
    "text": "releases, and ensuring our docs and"
  },
  {
    "startMs": 818880,
    "text": "documents are of extremely high quality."
  },
  {
    "startMs": 821600,
    "text": "We still do write a lot of docu comments"
  },
  {
    "startMs": 824160,
    "text": "ourselves, but it's very easy to give"
  },
  {
    "startMs": 826000,
    "text": "the agent a task of making sure"
  },
  {
    "startMs": 827760,
    "text": "everything is documented properly."
  },
  {
    "startMs": 829200,
    "text": "Everything has an example and everything"
  },
  {
    "startMs": 831279,
    "text": "actually is documenting the thing that"
  },
  {
    "startMs": 833680,
    "text": "it says in the way that it says. Um, as"
  },
  {
    "startMs": 836240,
    "text": "humans, you know, you'll go edit the"
  },
  {
    "startMs": 838000,
    "text": "code, but you won't edit the comment."
  },
  {
    "startMs": 839440,
    "text": "So, a lot of your comments will actually"
  },
  {
    "startMs": 841120,
    "text": "be out of date over time and things get"
  },
  {
    "startMs": 842800,
    "text": "very confusing. Um, and if you just look"
  },
  {
    "startMs": 845920,
    "text": "at the numbers, we've shipped more patch"
  },
  {
    "startMs": 847600,
    "text": "releases in our most recent DAXis"
  },
  {
    "startMs": 849120,
    "text": "version than we ever had before. So,"
  },
  {
    "startMs": 850880,
    "text": "we've been able to maintain weekly or"
  },
  {
    "startMs": 852880,
    "text": "multiple times a week release cadence"
  },
  {
    "startMs": 854560,
    "text": "for a large ambitious piece of software"
  },
  {
    "startMs": 857199,
    "text": "in a way that we would be scared to do a"
  },
  {
    "startMs": 858880,
    "text": "release earlier."
  },
  {
    "startMs": 861440,
    "text": "Um, one thing I'm not 100% convinced"
  },
  {
    "startMs": 864959,
    "text": "yet, uh, we have found varying levels of"
  },
  {
    "startMs": 867839,
    "text": "success is using AI to write tests or at"
  },
  {
    "startMs": 870800,
    "text": "least blindly writing tests. Um, one"
  },
  {
    "startMs": 874320,
    "text": "place we've struggled with Diosis is"
  },
  {
    "startMs": 876160,
    "text": "testing. It can be very hard to test"
  },
  {
    "startMs": 878399,
    "text": "foundational software, especially like"
  },
  {
    "startMs": 880560,
    "text": "end to end for complex systems. It's"
  },
  {
    "startMs": 883199,
    "text": "hard to test that your extension"
  },
  {
    "startMs": 884639,
    "text": "installs into zed and works the way you"
  },
  {
    "startMs": 886480,
    "text": "want it to do with literally opening zed"
  },
  {
    "startMs": 888800,
    "text": "and like using the extension. Um, the"
  },
  {
    "startMs": 892160,
    "text": "coding agents struggle here too to an"
  },
  {
    "startMs": 894480,
    "text": "extent. Uh, they also are, you know,"
  },
  {
    "startMs": 898000,
    "text": "have a tendency to write kind of sloppy"
  },
  {
    "startMs": 899680,
    "text": "tests. you'll give it a constructor and"
  },
  {
    "startMs": 901120,
    "text": "then it will go test the constructor and"
  },
  {
    "startMs": 902560,
    "text": "that's not a very interesting test. Um"
  },
  {
    "startMs": 905440,
    "text": "they can easily write tests for any"
  },
  {
    "startMs": 907279,
    "text": "given API but much like humans they fail"
  },
  {
    "startMs": 909760,
    "text": "to write the right tests. So we still"
  },
  {
    "startMs": 912880,
    "text": "find ourselves enumerating test"
  },
  {
    "startMs": 914320,
    "text": "conditions manually um crafting test"
  },
  {
    "startMs": 916880,
    "text": "APIs our ourselves and handling test"
  },
  {
    "startMs": 919920,
    "text": "runners um but it is sometimes a great"
  },
  {
    "startMs": 922639,
    "text": "sounding board to come up with the test"
  },
  {
    "startMs": 924800,
    "text": "ideas for a particular thing you're"
  },
  {
    "startMs": 926399,
    "text": "trying to to make sure has coverage and"
  },
  {
    "startMs": 928399,
    "text": "then enumerating the the edge conditions"
  },
  {
    "startMs": 931199,
    "text": "um but one place that we have actually"
  },
  {
    "startMs": 932880,
    "text": "really enjoyed using coding agents to do"
  },
  {
    "startMs": 934480,
    "text": "testing is building test harnesses. So"
  },
  {
    "startMs": 937519,
    "text": "fuzzing is a critical part of building"
  },
  {
    "startMs": 940720,
    "text": "like production grade software which"
  },
  {
    "startMs": 943120,
    "text": "means taking your application and"
  },
  {
    "startMs": 944639,
    "text": "putting it under millions of different"
  },
  {
    "startMs": 946639,
    "text": "inputs and quite often adversarial"
  },
  {
    "startMs": 949360,
    "text": "inputs basically like malformed inputs"
  },
  {
    "startMs": 952000,
    "text": "or uh ways of using the software that"
  },
  {
    "startMs": 954959,
    "text": "users should not be using the software"
  },
  {
    "startMs": 956399,
    "text": "but they they can use the software and"
  },
  {
    "startMs": 958560,
    "text": "coding agents are excellent at building"
  },
  {
    "startMs": 960240,
    "text": "these harnesses."
  },
  {
    "startMs": 963279,
    "text": "Um, one thing we've found that code"
  },
  {
    "startMs": 966320,
    "text": "extra code architecture is still an art."
  },
  {
    "startMs": 968959,
    "text": "Um,"
  },
  {
    "startMs": 971040,
    "text": "coding agents enable you to ship at an"
  },
  {
    "startMs": 973199,
    "text": "exceptionally high velocity. I mentioned"
  },
  {
    "startMs": 974720,
    "text": "this earlier. If the substrate on which"
  },
  {
    "startMs": 976720,
    "text": "your coding agents code lands is bad,"
  },
  {
    "startMs": 979839,
    "text": "their contributions will be bad as well."
  },
  {
    "startMs": 982079,
    "text": "Unlike a human engineer, coding agents"
  },
  {
    "startMs": 983920,
    "text": "aren't typically afraid to voluntarily"
  },
  {
    "startMs": 986160,
    "text": "go on a huge refactor of a system or"
  },
  {
    "startMs": 988000,
    "text": "redesign the architecture when a feature"
  },
  {
    "startMs": 989600,
    "text": "doesn't quite fit. They'll typically"
  },
  {
    "startMs": 991279,
    "text": "just ship. Most of our development time"
  },
  {
    "startMs": 993759,
    "text": "is actually now spent thinking about"
  },
  {
    "startMs": 995440,
    "text": "software architecture about what"
  },
  {
    "startMs": 997199,
    "text": "features we'll want in the future and"
  },
  {
    "startMs": 998560,
    "text": "how the system will evolve. Just like"
  },
  {
    "startMs": 1000639,
    "text": "human engineers can write spaghetti"
  },
  {
    "startMs": 1002480,
    "text": "code, so can the agents, but now just"
  },
  {
    "startMs": 1004880,
    "text": "faster. However, I will say with fable"
  },
  {
    "startMs": 1008160,
    "text": "level tools, the actual code quality"
  },
  {
    "startMs": 1010079,
    "text": "itself is so high, provided you properly"
  },
  {
    "startMs": 1012959,
    "text": "communicate your intent, that proper"
  },
  {
    "startMs": 1015279,
    "text": "software architecture will probably take"
  },
  {
    "startMs": 1017600,
    "text": "the vast majority of time in the future."
  },
  {
    "startMs": 1019600,
    "text": "Actual code writing, not so much."
  },
  {
    "startMs": 1022880,
    "text": "Um, one thing we do for Daxis, which"
  },
  {
    "startMs": 1025280,
    "text": "maybe you guys still do, maybe you"
  },
  {
    "startMs": 1027199,
    "text": "don't, uh, is we review every PR line by"
  },
  {
    "startMs": 1030319,
    "text": "line. Um, we definitely use AI review to"
  },
  {
    "startMs": 1033919,
    "text": "spot bugs ahead of time, but we still do"
  },
  {
    "startMs": 1036480,
    "text": "like to read the code that we ship. We"
  },
  {
    "startMs": 1039360,
    "text": "receive lots and lots of PRs from"
  },
  {
    "startMs": 1041120,
    "text": "strangers. Actually, uh, Daxis is big"
  },
  {
    "startMs": 1043280,
    "text": "open source project. Uh, and not every"
  },
  {
    "startMs": 1045520,
    "text": "PR is made the same. Uh we find that"
  },
  {
    "startMs": 1048000,
    "text": "users can be quite bad at communicating"
  },
  {
    "startMs": 1049760,
    "text": "their intent to the models. Contributors"
  },
  {
    "startMs": 1052080,
    "text": "don't usually think deeply about how the"
  },
  {
    "startMs": 1053600,
    "text": "codebase should evolve over time. They"
  },
  {
    "startMs": 1055200,
    "text": "just want their bug fix or their feature"
  },
  {
    "startMs": 1056799,
    "text": "in. Uh and many solutions are glued in"
  },
  {
    "startMs": 1059200,
    "text": "place. So we're we're not quite at the"
  },
  {
    "startMs": 1061600,
    "text": "point where the coding agents can read"
  },
  {
    "startMs": 1063200,
    "text": "our minds. Uh and thus we're still"
  },
  {
    "startMs": 1065440,
    "text": "limited by the medium of text. And as"
  },
  {
    "startMs": 1068000,
    "text": "ridiculous as it sounds, prompt"
  },
  {
    "startMs": 1069280,
    "text": "engineering is quite real. The quality"
  },
  {
    "startMs": 1071120,
    "text": "of an implementation can be very much"
  },
  {
    "startMs": 1072880,
    "text": "dependent on the the prompt that you"
  },
  {
    "startMs": 1074880,
    "text": "give the model."
  },
  {
    "startMs": 1076720,
    "text": "But in a sense, nothing really has"
  },
  {
    "startMs": 1078720,
    "text": "changed. Reading code has always been"
  },
  {
    "startMs": 1080400,
    "text": "more important than writing code. Um"
  },
  {
    "startMs": 1082799,
    "text": "maybe not in the beginning, but"
  },
  {
    "startMs": 1084000,
    "text": "eventually as the project evolves, uh it"
  },
  {
    "startMs": 1086160,
    "text": "does."
  },
  {
    "startMs": 1087679,
    "text": "So my closing thoughts on on using"
  },
  {
    "startMs": 1090960,
    "text": "coding agents to build ambitious"
  },
  {
    "startMs": 1092400,
    "text": "software is that code is now cheap, but"
  },
  {
    "startMs": 1094880,
    "text": "quality is not. Um the the job of a"
  },
  {
    "startMs": 1098240,
    "text": "software engineer has never really been"
  },
  {
    "startMs": 1100480,
    "text": "about putting lines of code on the"
  },
  {
    "startMs": 1101840,
    "text": "screen. It's it's been about"
  },
  {
    "startMs": 1104160,
    "text": "architecting elegant solutions to"
  },
  {
    "startMs": 1106080,
    "text": "complex problems to thinking 10 steps"
  },
  {
    "startMs": 1108320,
    "text": "ahead about how a system will evolve"
  },
  {
    "startMs": 1110720,
    "text": "about retaining flexibility in the face"
  },
  {
    "startMs": 1112880,
    "text": "of changing requirements. These facts"
  },
  {
    "startMs": 1115679,
    "text": "have not changed and the bar for"
  },
  {
    "startMs": 1117840,
    "text": "software engineering is higher than"
  },
  {
    "startMs": 1119600,
    "text": "ever."
  },
  {
    "startMs": 1121679,
    "text": "Uh if you would like to work on the"
  },
  {
    "startMs": 1123440,
    "text": "tools of the next generation of"
  },
  {
    "startMs": 1124720,
    "text": "software, Cognition, the people who have"
  },
  {
    "startMs": 1126640,
    "text": "acquired Dioxis are hiring. Uh the Daxis"
  },
  {
    "startMs": 1129440,
    "text": "team joined Cognition to be part of the"
  },
  {
    "startMs": 1130880,
    "text": "future and hopefully you will too. Thank"
  },
  {
    "startMs": 1133360,
    "text": "you."
  },
  {
    "startMs": 1135127,
    "text": "[applause]"
  },
  {
    "startMs": 1150848,
    "text": ">> [music]"
  }
]