How I AI - Spec-driven development: The AI engineering workflow at Notion | Ryan Nystrom

Starting point is 00:00:00 One line that I've been putting in my prompts lately is I literally don't know what I'm doing here. You got to explain it like I'm a five-year-old. I didn't start with writing code. I didn't start with anything. I just started with an empty markdown document. I actually just opened up Whisper and just started yapping about how this feature should work. I gave the Yap session to Codex and was like, write a spec. I then opened up Codex again, pointed it at this spec file, and I said, build it.

Starting point is 00:00:25 And basically one-shoted this. I've been in software engineering for 20 plus years. We were writing these documents and we were sitting in meetings with other engineers debating the merits of one implementation versus another. And now no more waiting for the meeting, no more waiting for review. I'm not a CI expert, but I kind of know what I want. And so other folks were kind of like, can you just bring some of your like puppy dog energy to like CI and just see what we can do? Your AI, your agent is never going to complain when you ask it to do this five minutes before the meeting starts. It is more relaxing and it's more fun and I feel like I'm getting more done.

Starting point is 00:01:02 It's weird to have this win, win, win. They do the triangle and they're like, pick two and you're like, no, I'm going to pick all three. Give me the whole triangle. Give me the whole triangle. Welcome back to How IAI. I'm Claire Vow, product leader and AI obsessive here on a mission to help you build better with these new tools. Today we have Ryan Nystrom from Notion and he's going to show us as an engineering manager how you can never prep for a standup again. We're also going to see how you can get a background agent.

Starting point is 00:01:31 to write code for a fix your friend texted you, and how spectraven development really works in a codebase at scale. Let's get to it. This episode is brought to you by WorkOS. AI has already changed how we work. Tools are helping teams write better code, analyze customer data, and even handle support tickets automatically.

Starting point is 00:01:51 But there's a catch. These tools only work well when they have deep access to company systems. Your co-pilot needs to see your entire code base. Your chatbot needs to serve. search across internal docs. And for enterprise buyers, that raises serious security concerns. That's why these apps face intense IT scrutiny from day one. To pass, they need secure authentication, access controls, audit logs, the whole suite of enterprise features.

Starting point is 00:02:18 Building all that from scratch, it's a massive lift. That's where WorkOS comes in. WorkOS gives you drop-in APIs for enterprise features, so your app can become enterprise-ready and scale upmarket faster. upmarket faster. Think of it like Stripe for Enterprise features. Open AI, perplexity, and cursor are already using WorkOS to move faster and meet enterprise demands. Join them and hundreds of other industry leaders at WorkOS.com. Start building today. Ryan, welcome to How IAI. I am really excited because you're going to show us, I think, start, medium. advanced mode on some AI coding stuff.

Starting point is 00:03:04 And so before we jump in, how has AI just changed how you live your life at work? I mean, it has completely upended the way I work. Like I've been doing this. I did a lot of like mobile iOS work in my past. And I worked the same way every day for like 12 plus years. And then the last year I have changed IDs, terminals, tools, like whatever, like 10 plus times.

Starting point is 00:03:37 So it's like really weird and scary to be like changing this stuff so much. But it's also I feel that like I've been doing this for a while and I'm feeling like so much like joy and freshness and like newness in this that like I wake up every day super excited to like tinker and build things. And I'm also like working faster and harder than I feel like I ever have. but like in a good way. I know people are like kind of freaked out about like, everything's changing the pace is up. But like it's really energizing for me. Well, you are not alone.

Starting point is 00:04:09 I think every how I guess has come on and said, I'm having more fun. I'm working faster and everything is different. And what I love about what you're going to show us is it's not just the set of tools, which I think has changed or how we write code has changed. How you like run a team has changed. So, and I think for the better. You know, so I would love to see how you're using Notion AI to actually run teams differently.

Starting point is 00:04:36 Yeah, so for context, I manage a team of like six, seven people. I'm like an engineering manager or technical engineering tech lead manager or whatever we call it. I manage people and write code is like my role, which I love. And I've run a bunch of different projects here at Notion. the one I'm going to show today, we've nicknamed Afterburner. Like, kind of a quick backstory is I've been kind of vocal about our, like, DevX CI for a while. I've just, I worked at places where it's really slow. I've worked at places where it's really fast.

Starting point is 00:05:15 And I came here and I was like, we're kind of like in between. But I feel like we're like slower than we need to be. And eventually this like caught up to me. And somebody was like, could you just come fix it? and like come work on it. And I'm not like a infra expert. I'm not a CI expert. But I kind of know what I want.

Starting point is 00:05:36 And I think also most importantly, the group that I manage and the org that I work in, we're a little notorious for being like really fast and like very, very AI-pilled. And so other folks were kind of like, can you just bring some of your like puppy dog energy to like CI and just like see what we can do? And that's what we've done.

Starting point is 00:05:58 So we had this really aggressive goal to cut our CI into like a quarter of what it is. We're on the path to doing that. So what I want to show you is a little bit about how we run projects in Notion and how I'm using AI to kind of streamline those projects. So what we're looking at here is like our project hub called Afterburner. And so in here, I've got all this documentation. I've got databases. We'll look at some meetings. I have like an automation that looks for any sort of like little wins if we knock off seconds from different jobs or whatever.

Starting point is 00:06:34 And we keep it, we keep track of them in here. But what I want to show you is basically how we run our meetings. Our small group, we run a stand-up every single day. And doing stand-ups where everyone just like is kind of like dead-eyed and going around being like, I did this, I shipped this change, or, you know, no updates for me, thanks, is like painful. And in my opinion, like a huge waste of time. I want to like get to the meet. So we have this kind of like automated meeting template that shows up every single day, that we run our meetings, which is basically every day.

Starting point is 00:07:14 And it starts blank. And then what I have set up behind the scenes is a custom agent. So we chipped this Notion AI custom agent stuff a little bit ago. And this runs right after the meeting template gets generated. And it looks through all of our like Slack conversation in the last 24 hours, any tasks in Notion that we close, any pull requests that we've like merged, just like all sorts of contexts. Oh, and it looks at like yesterday's meeting transcript as well.

Starting point is 00:07:49 And then it compiles it basically like a pre-read. and this is one from a week or so ago. And it, yeah, it shows, it pulls metrics. It can show us, like, what our latest CI time is. It shows some of the things we've decided. It shows, like, progress on different, like, projects or, like, different things that we're trying to, like, make faster. Bugs, feedback, open questions, tat, like, anything that's, like, of concern.

Starting point is 00:08:18 And, like, I can be. basically work up until like the minute of our meeting without having done a bunch of like prep. And then we all get on a video call and we look at the screen and we're like, okay, here's what we need to talk about. And we'll like hit each bullet. We have like meeting notes that we run down here within Notion. And like all the context is basically like captured all of the like agenda is like set up for me. So we spend the entire time talking about like problems, decisions, wins, findings, like, what are we going to work on next? And it's less the like, oh, I did this thing.

Starting point is 00:09:00 Yeah. What I want to call out for folks that are maybe listening and not watching is this is a very detailed meeting, like kind of like pre-read slash status update, which if you're a, you know, TLM or an EM, running good meetings, unfortunately, is part of the job. And there's a really big difference between a good stand-up and a bad stand-up. And I think the ones you describe, these, like, wrote, like, I wrote this PR. Today, I'm going to start on this. And then basically, like, a notes document that's a very high level because some human is putting it together, it stinks. And you start to, like, do the other thing that I found, and maybe I'm curious if this has impacted how you work or just made it easier, which is like, you start to have those meetings less, because the updates aren't super rich.

Starting point is 00:09:52 People don't feel like it's a good use of their time. And I think you lose something by reducing the frequency. But if you can have high bandwidth, high quality meetings with high frequency without the overhead, I think you can get better, more detailed work done collaboratively, especially if you're running like a remote team or not everybody's in the same room. So I'm curious, do you feel like just? Just being able to get to this level of detail in your stand-up has improved how you can actually build this product and do this work, basically. I've been in way too many meetings where I can tell, like, everybody's eyes are glazed over.

Starting point is 00:10:30 Nobody's paying attention. And I, like, I have run enough projects at this point where, like, to me, that's, like, pretty dangerous. Because, like, everybody in that room has ideas. They have insight. But, like, because we've made it, like, so non-productive and, like, like not engaging that like we're not we're just like we need the point of the meeting is like exchange ideas and information and if we're like not doing that like literally what's the point the whole like this could have been an email like that's what the meme is from and so I I found

Starting point is 00:11:02 that this is basically a good conversation starter and a key thing to me that that tells me that this is working is when I'm like I'm running these meetings and kind of going through through like bullet by bullet and then I see like, oh, somebody like fixed our mock server environment in our, our test tests. And we're seeing a test improvement by like up to 13%. I'm like, oh, I'd miss that, you know, like that's super cool. Like tell, like, let's talk about it. And then from that maybe like, oh, there's additional headroom that we can make on this. And so all of a sudden, like, let's drill into it. And so just like, I also think it sort of democratizes that like sharing so that you've got some people that

Starting point is 00:11:49 I could talk for an entire 30 meeting without shutting up and then other engineers are like kept keeping to themselves but like brilliant and super talented and I think it's great. This kind of like raises everything up to the top and makes it very visible.

Starting point is 00:12:06 I was thinking last night truly this is a little bit of a sidebar but I was thinking last night how people don't know this it's very very confusing as Claire Vaux the podcast hosts. I am such an introvert. I'm a crazy introvert. I will avoid humans with every fiber of my being. And I find like AI as a proxy for me to get over my own anxiety communicating with the other. It's like so insane. But I really

Starting point is 00:12:32 feel it whenever I like message my open claw and I'm like, hey, Polly, can you email so and so blah, blah, blah, blah on my behalf? My anxiety is gone. But when I sit in front of my Gmail and start typing an email, my introverts start showing. I'm like, I can't, I can't do it. So I think this is a real point, which is like you got to pull information out of everybody. Yeah. And there's like spikes and valleys of who is going to come and show off their work, how detailed they're going to be. It does not mean the work is not good. It does not mean they are not talented. They just have different skills in different environments. And so I love this call out of like, we're going to pull equal amounts of information out of everybody because the playing field is, you know,

Starting point is 00:13:15 is equal. I also want to go back to something you said, which makes me think about burnout, but she said I can work up until this meeting starts. And I know so many people who feel like they're in meeting after meeting after meeting. One, because when they're not in meetings, they're preparing for meetings. So you're getting rid of that. Yeah.

Starting point is 00:13:35 And then two, I think managers being able to, for example, code up until the stand up until the standup is such a burning. now like protection mechanism, which is you would much rather your managers hands on the code, hands on doing work, filling a creative impulse, then prepping for meetings. Sure, they're great at running the meetings, but let's do that in like this just in time mechanism where they're supported through these automations. And so I just imagine, I used to feel so stressed as a leader and as a manager being like, if I'm not in a meeting, I'm prepping for a meeting and then I'm in a meeting. And if I can carve back more of my time to just do real things, that feels so much, so much, so much better. So I'm guessing you're having more fun just through the balance of the

Starting point is 00:14:25 kind of work you're trying to do. I'm, yeah, I'm having so much more fun working this way. I have done the like run a big engineering group where yes, you're spending half of your time just compiling information, synthesizing it, writing reports, like doing all the stuff. and I hated it. Like, it's so draining. And now I feel like I'm in a sweet spot where I can support and work with a team of like very talented individuals, but also not have to, yeah, be like doing like paperwork the entire time. Like, yeah. I don't want to do like the TDM.

Starting point is 00:15:04 I get a lot of joy out of working with people. It's like why I like managing. I like talking with them, having fun. I love like solving hard problems together. And then I like building stuff. And, yeah, I think that we're maybe at an inflection point where, like, maybe this is controversial or not. But, like, if you're, like, a line manager, like, write, right code, you know, get in there. And, like, stay close.

Starting point is 00:15:34 Maybe don't do the P0 hero projects. But, like, help your team fix bugs, like, make optimizations, like, whatever, you know. I, it's just, it's so easy now. I mean, I'm going to pull that thread all the way up, which is like, directors of engineering, VP's of engineering, like CTOs, CPO's. Yeah. CPO's. It, right, write some code. Now is the time. And I say this all the time on this podcast.

Starting point is 00:15:59 This is the era of the hard skill. This is not, how do we get better at my, my soft skills, and managing stakeholders? This is literally like, how do you write code? How do you write automations? How do you learn these new tools? How do you understand what models do. do what for your own skills? That I think is super important. Okay, we could go on forever on this topic.

Starting point is 00:16:17 I do want to show folks how you built this. Yes. Because we haven't seen, actually haven't seen, we've had a couple of Notion folks on the podcast. We haven't seen Notion AI in action. And I just want to see kind of your thought process on how you build something like this out. So I'm going to flip over to this is our custom agent.

Starting point is 00:16:36 Don't ask me why we got this like potato theme for like the entire project. I think it was kind of something about like, like CI is just this like cobbled together like mess. And so we're going to like make the potato like a rocket ship. I don't know. I don't even know if that makes any sense. But like now we're having fun with it and we have reactions and like the agents. And it has spun off into its own thing, which is fun.

Starting point is 00:16:58 But this is our hot potato agent. So you can see I have this setup to run at 9 a.m. every single day. It's also set up for chat. It's set up if the agents mentioned. But we never really use any of that. And probably the important. part is the actual instructions. So in this instruction page, giving context on like what the purpose of this agent is, I am telling it to run, yeah, look back at 24 hours. So basically telling it your job is to run every single day. And I only want you to look back for the last like 24 hours of activity. I'm explicitly telling it to use subagents, which is kind of a sleeper feature in Notion AI. Like this exists, but we don't, you don't really, push it to use it very often yet because it's one it's very expensive and two it can be kind of

Starting point is 00:17:46 finicky sometimes but um i help build it so i know how this works uh and then i ask it to kind of like fan out and do like a map reduce where i'm saying go use the honeycomb mcp to figure out what the latest metric is look in our project channel and like find updates feedback questions i'd tell it where the task databases and how to look for tasks within this project and then how to find yesterday's meeting. And then I give it a template in the instructions or I'm like, this is your format. I care about CI speed decisions, progress, changes, bugs, questions, risks, a little bit of guidance on writing. And then when it's done, I have it posts to Slack and I like emphasize this like I want it to be brief and fun.

Starting point is 00:18:36 And sometimes it's really corny. And then sometimes it's like really good. And it'll be very quirky and just post this like link in our Slack channel. And it's like, hey, here's your pre-read. Some little quibble about like whatever, you know, like, hey, you guys are not making enough progress. And then that's it. And then we have our meeting note. It's updated.

Starting point is 00:18:59 What I like the most about this one, you can show you some of our like internal settings. So this is like I give it access to all these things. I'm like, you can only view all of this stuff because I don't want it going and like modifying our task database or project database. Like everybody at Notion uses those. But this meeting's database in particular, I'm like, you know, you can edit content because this is the one you're going to like write and update the page. It can read from our Slack channels, respond to our project one. And then this was new to me. actually when I set this agent up is our MCP.

Starting point is 00:19:36 So we've had MCP and we have this other thing called Workers, which is kind of like writing code. I haven't used them very much within custom agents, but in this one in particular, I'm like, I know exactly where this metric is. It's in Honeycomb. And so I like just configured the MCP in Notion. By the way, I like used the agent to like set itself up.

Starting point is 00:20:00 I was like, here's the query. I literally gave it a screenshot of the Hanging Home Query. I was like, I don't know how this works. Can you just like update your instructions? I love that you screenshot it. You didn't even copy and paste it. You're like, please OCR this screenshot. Exactly. Too lazy.

Starting point is 00:20:15 I'm like, here it is. Just take it. Figure it out. And it kind of, it got it mostly, like most of the way there. I had to fiddle with it a little bit. But yeah. What I appreciate about this, and again, for anybody trying to just brainstorm

Starting point is 00:20:30 workflows where AI can actually have a huge impact on your productivity at work or life. I just feel like write down what you would do if you had time. If you had time every morning at 9, you would sit down and you with your eyebrows would go through Slack. You would go through Honeycomb. You would ask people what's going on. You would look at GitHub and then you would compile it. And then you would try to be very fun in Slack.

Starting point is 00:20:53 Like it's just a description of what you would do. And it doesn't have to be that complicated. You can iterate so quickly on it. What I appreciate about this, you know, versus old era of more deterministic, like, workflow-style builders is the updates are so easy to make. Just change the natural language, redo the order, change the trigger, give it more access to data. And then it's ready to go. Yeah. And you know, I think the other thing I've gotten hung up on when.

Starting point is 00:21:29 trying to think about like these automations and I've seen others do is that like you get this like you start gigabraining it and you're like well how am I going to save like five hours of work a day and I what you just said made me realize too that like the tedium that this removes for me is not like world changing but it's like 20 minutes a day and that's like 20 minutes I can spend doing other stuff And that like, it's not even just about like saving that 20 minutes, but it's like protecting my brain from like having to context shift about all this stuff and like ingest it. And instead, yeah, it's just I know that the information will be there when I'm ready to like read it and I'm ready to like shift gears to this project. Um, rather than yeah. I hate doing the like read this update, copy all this information like put it into it like an update board. Like it's soul sucking. I mean, you and I have been doing this for a while. That's like that was, I feel like 70% of my job.

Starting point is 00:22:32 At some point, 70% of my job was just like, what's going on? And how do I massage it into a format appropriate for the audience at hand? And it's always the same information. It's just like, what's the executive version of it? And what's the team version of it? And what's the full team version of it? And like, oh, like my shoulders drop out of my ears when I realize we don't have, like we just don't have to do it anymore. And the other thing that I think people maybe underappreciate about AI and this like just in time delivery is your AI, your agent is never going to complain when you ask it to do this five minutes before the meeting starts.

Starting point is 00:23:11 I know. It's so great. It's so great. It's just like when you have it, drop it, get it done and out of your brain. I think is, again, I go back to like burnout and enjoying your work and reducing toil. And it just feels like. a more relaxed way to work. Yeah. We're happy.

Starting point is 00:23:33 It's so funny because it is more relaxing and it's more fun and I feel like I'm getting more done. Yeah. It's weird to have like this like win, win win. Yeah. You know, they do the triangle and they're like, pick two and you're like, no, I'm going to pick all three. Yeah. And I'm like, give me the whole triangle. Give me the whole triangle. That, stamp it for the YouTube thumbnail. Give me the whole triangle. All right. This episode is brought to you by Orcas, the company behind open source conductor, which powers complex workflows and process orchestration for modern enterprise apps in agentic workflows. Legacy business process automation tools are breaking down. Siloed low-code platforms, outdated process management systems, and disconnected API management tools

Starting point is 00:24:13 weren't built for today's AI-powered world. Orcus changes that. With Orcus Conductor, you get a modern orchestration layer that scales with high reliability and brings humans, AI, and systems together in real time. It's not just about tasks. It's about orchestrating everything. APIs, microservices, data pipelines, human-in-loop actions, and even autonomous agents. So build, test, and debug complex workflows with ease, all while maintaining enterprise-grade security, compliance, and observability. Orcus. Orcestrate the future of work. Learn more and start building at orcice.io. Let's talk about, so we're talking about how meetings happen. Love this. You do write code, though.

Starting point is 00:24:57 I do write code. Sometimes with your fingers and sometimes with, you're my favorite harness of the moment. So let's go to how you get coded. Yeah, I want to show the little bit of a new workflow that we have going on in Notion. I honestly don't think this is necessarily a big feature that we're going to ship or we might ship some version of this feature.

Starting point is 00:25:22 So this is all basically internal only at this point. But prior to this, the way one, I love Codex. I've been a Codex stand for like, I don't know, six, seven months now. And we started building this like Codex integration into Notion. Prior to this, it was like, I mean, obviously you're using the CLI to like write your prompt. And then they created the Codex app, which is nice. But I'm still like writing my prompt in this thing. So I started actually writing prompts in Notion.

Starting point is 00:25:55 pages where I can be a little bit more like freeform and structured and you know I'm in like the I don't have to worry about like hitting enter and then like oh shit like I sent my prompt uh I can actually like write a document in a notion but then of course I'm like cop or like highlighting all of the texts copying it going into my terminal hitting paste and like letting it go it's like fine um there's like MCPs and other stuff that I could be using but I'm I'm a very simple person It was like too much, too much stuff. So we built this thing that we're kind of like calling it, I think we're calling it both Software Factory,

Starting point is 00:26:33 but I like its internal project name is Boxy, because it's like all these little VMs that we install Codex and ClaudeCode on. It's our little boxes, where now we can actually invoke them from like tasks within Notion. And so literally happened this morning, a friend of mine who's a Notion fan, texts me, he's like, hey, I like the tab block that you built, which is this thing.

Starting point is 00:27:00 He's like, but I really wish I could like, you know, I can like copy, I'll show you now. If I click on the dot, dot thing next to a block, there's like copy link to block. He was like, man, I really wish I could just like copy link to a tab and then like send it to somebody. And I was like, oh yeah, that sounds really easy. So I opened up this task.

Starting point is 00:27:21 And I just took some notes and I dropped in this screenshot showing them where it could live. You know, this is like on the tab block. If I right click on it, we get this little flyover menu. And I just described the task. I was like, yeah, let's put a copy link button here. I also noticed that hovering over the delete button didn't change to red. So I was like, yeah, we should fix that. And I was like, I know one edge case is like, you're going to have to like,

Starting point is 00:27:48 if you land on a URL where we're like, we're like, linking to this block and this tab like it's got to like switch to the tab on like fresh refreshes so i like kind of outlined all the cases but you can see here this is one two three paragraphs four like four sentences and a screenshot it was like not not a lot and then uh this new thing that we built i can actually mention codex from within our comments and this triggers are like boxy uh this is all our like internal dev tooling stuff. And then it got to work. And I was looking at the timestamps earlier,

Starting point is 00:28:26 and I think 1040, 1051 started the implementation. And then another 10 minutes later, it replies with a pull request link and a preview URL because we do the like preview environment stuff. And it had like built the entire thing. And if I switch actually over to here. the poll requests. It built like the entire thing. It was like, here's how I tested it. And this was actually the coolest part to me is like it actually uploaded screenshots of it

Starting point is 00:28:58 doing its own like UI verification. And there was like a CI failure in it. So I was like, hey, I replied down here. Like, oh yeah, this part of this code change. I was like, I don't know what is going on here. This doesn't make sense. And there's some type check things. And it was like, cool. here's why we did this change and I fixed your types and then there was a merge conflict let's also talk about just a world new world

Starting point is 00:29:27 the emotions around code review like I love that you're like I don't get this not I'm going to try to get it or I've done my best to investigate and I'm pretty sure this is wrong like literally just I don't get it I don't get it

Starting point is 00:29:43 fixed I mean I didn't like sending that to somebody one of the team mates be like no look it up it makes it so that's one of my one of my code review mechanisms as well as I'm just like I don't think this is right I yeah one

Starting point is 00:30:00 one one line that I'm been putting in my prompts lately is I'm like I literally don't know what I'm doing here you need to explain this to me especially doing all the CI stuff I'm like I'm in over my head like you gotta you gotta like explain it like I'm a five year old

Starting point is 00:30:16 And sometimes it's like, literally like caveman style. Like here are the things. And I'm like, I needed this. Well, what I appreciate about this. And we'll talk about it in a little bit later, I think, is we're not getting any of these ClaudeCode warm fuzzies from Codex. I love it. But when it talks to me, I'm like, I feel real dumb when you're talking to me codex.

Starting point is 00:30:36 Can you? It's like, okay, okay, dumb, dumb. Here's what's going on. So I think that sort of like, I don't understand what I'm doing. please explain it to me is I love it, but I think it is a codec specific experience. Claude shows up and is like, hey, buddy, guess what I made for you? Okay, so stepping back before we go too deep on the personalities of all these coding models, you have built just at mention.

Starting point is 00:31:07 And I think people miss this, but it's important for particularly our engineering team and engineering leader listeners to pay attention to, which is, Everybody that knows that's smart. It has some background agent hooked up to a virtual machine that they kick off so that you're not spinning up your local environment and doing all this stuff on your machine. And it just, I think the, we'll talk about this, but the velocity plus your devX, plus your CI, I think is a huge piece of AI adoption in engineering. So if you are not, if you don't have a like VM strategy and background agent strategy in

Starting point is 00:31:44 or large engineering org. Time to get one. Time to do it. And we'll just ask Codex to build it. I don't know what I'm talking about. But I think this is, Claire said this is a good idea. Please build. Make no mistake.

Starting point is 00:31:59 Okay. Last one, we have a last use case. You were going to walk us through. Yeah. So we recently rebuilt our entire agent harness. The quick TLDR is. just like everybody else, we reached like this point of tool and instruction fatigue, where you had this bloated system prompts.

Starting point is 00:32:20 So we're like, okay, we need to dumb this down. And we borrowed the kind of like skills and progressive disclosure concept from coding agents and brought that to Notion AI. And when we were doing this big rewrite, we were asking ourselves like, you know, it had only been like six months since we shipped the last rewrite. And this time we were like, what would we do differently? Like seeing the state of the art with coding agents. and someone had this really great idea to, like, let's not start with code.

Starting point is 00:32:50 Like, let's just start with specs. And what we've ended up building is we have this, like, in our checked into our code base. You see this, we have this, we're looking at, this is a Notion repo, but we have this agent specs subfolder. And within this subfolder, we have all of these markdown documents. This is one that I worked on. We have this thing in our AI called Ask Mode, where we basically ban all the mutating tools. So it can only just like read and answer questions.

Starting point is 00:33:21 And so when I was building this, I didn't start with writing code. I didn't start with anything. I just started with an empty markdown document. And I mean, I actually just opened up like Whisper and just started yapping about how this feature should work. And at the end of it, I gave that. I gave the Yap session to Codex and was like, here's our other like spec library, learn the format, take my information, write a spec. And then it spiked the first version. I did a couple revisions on it.

Starting point is 00:33:56 And I ended up with this markdown document. Now, the markdown document is like, it's nice. But what we did with it next, in my opinion, is I kind of think that this is like the future of software engineering, where I then opened up. Codex again pointed it at this spec file and I said, build it. And it basically one-shotted this because the entire spec file is so comprehensive with code pointers, with down at the bottom, we have verification. It was like, here is how you're going to verify all of this stuff works. And we've even built our own CLI tools so that you can run Notion AI from the CLI.

Starting point is 00:34:36 and it could, once it's done seeing that all the tests pass, it can actually just spin up Notion AI itself, send it queries, send it questions, enable ask mode, and then see the transcripts and like see what actually happens. And I think the first shot of this took a couple hours, but I came back to whatever, a couple thousand lines, did some code review, played with it myself, and I was like, it's right, it's like done.

Starting point is 00:35:03 And, you know, since then we've made, I can, the beauty of this is like, this is inversion control. So I can go to the past changes of this spec file. And I can see how the spec has evolved. And I could go look through all of the code changes, which also have their own, like, history. But this is now the sort of like source of truth for how this part of notion AI works. And it's just in plain English that can then be verified and implemented by agents. Well, I think the other thing that people don't appreciate is, you know,

Starting point is 00:35:36 taking this outside of the engineering flow is this plain English can be ingested by other parts of the business that need this information. So let's say you need to then release this feature via some sort of marketing. This is actually like a pretty good asset that explains how it works. That can be translated into another thing in a way that like code itself is still a little and tractable. And so this idea of spec-driven development, but what I like about what you said, I don't want people to miss is the way you make these updates is you update the spec and go, go look, make the update, change the thing. And so like the spec is the source of truth. The spec as the change log, I think is a really interesting model. And for people that aren't

Starting point is 00:36:26 watching, it's very detailed. It's very technical. So it's not like there is not code in the spec. There's just not all the code in the spec. And I think that's a really kind of good hybrid model for experienced engineers to start to bridge into what would it look like to have an agent do more of your coding work while you still do architecture work, while you still do design work, while you still make sure that the thing is going to scale. Yeah, exactly. I view our job as like engineers evolving into like systems thinkers and architects and not even just necessarily writing like the spec and thinking about the behaviors but most importantly is like the verification loop like is it like how should it verify correctness of this feature or this

Starting point is 00:37:20 change and honestly it's like if it can't or if like the verification's a little hazy it's like that's the first thing you actually should be going and doing it's like do you have a tool to let the agent run itself. That's like one of the first things we did with this project was like, we should actually build a CLI so that I can tell Codex, like, send this prompt and like see what happens. And now that we have that, then we can take these specs and actually just like go deeper and deeper and deeper. And so it's like we're still doing engineering, but I'm not doing the like plumbing work of like wiring up this ask mode feature. Well, and what I think is really funny is I've been in software engineering for 20 plus years. Like we were writing these. We were writing

Starting point is 00:38:00 documents anyways. We were writing technical design documents. Exactly. Document documents anyway. And we were sitting in meetings with other engineers debating the merits of one implementation versus another. And then we still had to go write the code. Yep. It's like really, it hasn't added work to go into this model. It's maybe like shifted the emphasis of where the human attention goes. But these were documents that at least I was in every org that I've ever been in writing before. And the other thing that I think has changed so much is those docs then waited for review. And they waited for a meeting. And now, no more waiting for the meeting, no more waiting for review, ship it, have a verification loop, debate it on the

Starting point is 00:38:48 merits of it being live and working versus the theoretical merits of it sitting in a document, waiting for everybody's calendar to open up for, you know, a live argument. Yep, yep. Couldn't agree more. Let's do it. This is my, I just you and I could talk all day about this. Just recap for everybody because I know I got to get you out of here. Three use cases. One, never prep for a meeting again.

Starting point is 00:39:12 Hook it up not only to your Slack, but to your meeting notes, your GitHub, your telemetry, and build the best stand-up meeting. So no one has to stand there, glassy-eyed, being boring, giving updates. second use case, background agents, at mention from wherever you work, Notion being a great place to do that, kicking off virtual machines, getting PR's done, just saying yes when your friend texts you, can you ship this feature? And then the last one, putting all your specs in your repo, using them as a source of truth for a more autonomous coding agent like Codex, let it cook for a couple hours,

Starting point is 00:39:48 review the code, and then when you update, update the specs, don't update the code. Did I get it right? Nailed it. Okay, let's do a couple lightning around questions, and I'll get you out of here. You and I love Codex. Why you love Codex? I'll tell you why I love Codex, but you go first. Okay.

Starting point is 00:40:03 Well, so I first fell in love. Oh my God, did I just say that? You did? I first fell in love with Codex because when I was like evaluating both like Claude Code and Codex, I found anytime CloudCode like filled up its context when, window, it would just kind of like lose the plot really quickly. And codex, I don't, I don't know exactly what it's doing. If it's the model, if it's the compaction, if it's like both, they can, it can grind for like hours. And with with the way that I work, both the systems and things that I work on,

Starting point is 00:40:43 and just like, I like to be able to fire off a bunch of them like at the same time and then like go to a meeting or go do something else. Or like, like, uh, spend my time kind of like round robin managing like all of these agents like I I don't necessarily want to I'm not the person that is like sitting with the browser open and like um some agent next to it and like iterate look at the browser iterate look at the browser I'm like the more the closer I can get to one shotting solutions the better because that frees me up to do other stuff um so I found that like codex was was pretty good about that I also just feel it's like pretty simple like it's There's not a lot of bells and whistles.

Starting point is 00:41:24 It's not too fancy necessarily. I'm happy with the addition of like MCP and skills and some like other stuff coming out. I also really love GPD 5-4. I think it's a great model. So I'm like all of those things together, it's just it really like matches my working style and type of work a lot. Yeah, I'll tell you why I love Codex. And I sent this to somebody I said, work trees everywhere. ports 3,000 through

Starting point is 00:41:54 2009 spoken for like we're just we're just going across across the board I do think it's fit at long running tasks I like its concept of projects because I run a lot of different projects I think it's just like a very helpful mental model and then it's great

Starting point is 00:42:10 at code review honestly oh yeah it's just like a really good code reviewer it's a really good security reviewer tireless uncomplaining with the most tedious of things. And so I find it's one of my daily drivers. I just I really like it. It's good. Okay. Second question because you and I also agree on this. Give people the reason. This isn't even a question. This is a demand for you to share my point of view, which is why bang on developer

Starting point is 00:42:41 experience and CI speed right now? A couple of things. To me, CI was like super important prior to agents. Because in my opinion, I mean, it's not even an opinion. It's like fact that like the faster your CI completes, the faster you get signal, the faster you're like the more

Starting point is 00:43:02 comfortable your engineers will feel with like making changes and doing things because they know like I can make a change and get it pushed into dev or into production really quick because I know my CI's fast. If it's slower, then you're building up these like monster changes and you're going to

Starting point is 00:43:18 to be even slower about like judiciously like reviewing every little like tiny thing when I'm like a learn through like doing sort of person and if I can close the iteration loop like even tighter then I'm going to be putting out a change people are going to be using it I will take that feedback I will make another change and I'll have to worry about um you know taking another day before like the deploy train is ready to go. I just want to like, I want to crank really, really fast. And that was all pre-agent. And now that we're like in the agent land, it's, it's that, but like on steroids because agents don't get tired. They can work on a VM. They can work like while I'm sleeping. And, you know, if I'm take, if I've got a CI loop that takes an hour to run, that your agent's just going to

Starting point is 00:44:09 sit there and spin for like an hour waiting for results to like do something. If it takes three minutes to run. Like, holy crap, how much more stuff are you, you as a human? And then especially as your, like, little swarm of agents going to be able to get done. Like, so, so much more. And so super important. I agree. And we just had Steve from Stripe on. And they're doing like, 1,300 agent PRs a week. Oh, my God. You cannot do. Yeah. You like cannot do that if your CI is slow. It's just, you might as well be throwing all those PRs in the trash. And so I do think there is just a true mathematical limit on your capacity to ship code to production, that is a reflection of how fast your CI pipeline is. And so if you are, again, engineering leaders,

Starting point is 00:44:56 if you are not spending time on making that fast and you are tolerating something slow there, you're not going to get the benefits of AI that you could. And honestly, like, good for AI, good for humans. No engineer wants to sit waiting for their stuff to hit production. It's miserable on the engineer side, lots of downsides. So allocate time to your pipeline, please. Yep, yep. Ryan and Claire say so. All right, last question. When AI is not listening? It's writing bad specs. It's not being funny in Slack. What is your prompting strategy? Do you yell? Yeah. Yeah, I can tell. I can be a little bit of a diva sometimes if it like really goes off the The other prompting strategy that has like saved my ass working on the CI stuff lately is because like even the best models I feel like can sometimes be a little sycophantic.

Starting point is 00:45:54 I'll be like, I will just like be like you're wrong. Like you need to defend your argument because I like I want it to defend it in the way that I like I want. But I'm like I just need I just need to see the evidence that if I push counter to what it has done that it can. can back up with like good pointed reasons rather than just be like, are, you know, are you sure this like change looks okay? It's like, oh boy, it's like totally fine. I'm like, no, no, no. I need like the cited hard argument against it because like in a lot of times, like with the CI stuff. I'm like, I don't know what I'm doing. I know generally what I'm doing. But like the specifics are a lot more nuance and I like, I need to get this right. And that's been super helpful. And then when it goes

Starting point is 00:46:40 off the rails, I'm like, I can be such a divot to it. Escape, all caps, no. That's what I do. Interrupt and no. Steer the conversation. A little arrow button has never gotten so much work. Ryan, this was so much fun. Where can we find you and how can we be helpful? You can find me on X.

Starting point is 00:47:05 I'm out there shilling for Notion right now a lot. But that's basically where I spend my time. Yeah, my DMs are open. So if you have notion problems, I like to try and fix them. I'll send it to Boxi and we'll get it done. Perfect. Love it. Well, thanks for joining the podcast.

Starting point is 00:47:22 Thanks for having me, Claire. Thanks so much for watching. If you enjoyed this show, please like and subscribe here on YouTube or even better, leave us a comment with your thoughts. You can also find this podcast on Apple Podcasts, Spotify, or your favorite podcast app. Please consider leaving us a rating and review, which will help others find the show. You can see all our episodes and learn more about the show at how IAIIPod.com. See you next time.

How I AI - Spec-driven development: The AI engineering workflow at Notion | Ryan Nystrom

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.