Successfully coding with AI in large enterprises: Centralized rules, workflows for tech debt, and training your team | Zach Davis (Director of Engineering at LaunchDarkly)

[00:00] Vibe coding is not an acceptable enterprise development strategy. I love it. I can do 100 commits a week by myself on my side project on my startup. But when you're working on a code base and a platform like LaunchDarkly that powers trillions and trillions of experiences every day, you can't take the same strategies and tactics that a vibe coder could take. One of the things that I realized is what's good for humans is also good for LLMs. And so I really [00:30] The repo is well set up for humans to know how to work in it. So we have front end organization, we have accessibility, we have a JS style guide. So all is this very detailed documentation that we've put into the repo itself rather than have it in other places. And this way, LLMs can access it, humans can access it, et cetera. I think all the engineers out there are like crossing their fingers and hoping that there's one rules protocol to rule them all that shows up. And I think what you've shown is you can just create that yourself and then that makes it much more scalable. [01:00] Welcome back to How I AI. I'm Claire Vo, product leader and AI obsessive here on a mission to help you build better with these new tools. Today, we have a great episode for anybody trying to deploy AI agents in a real engineering team with a real code base, not just Vibe Coding. We have Zach Davis, Director of Engineering at LaunchDarkly, who's going to show us [01:24] how he sets up centralized rules and docs for all his AI agents, [01:28] uses AI to burn down tech debt and keep his hiring bar high.

[01:33] Let's get to it. [01:34] 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. [01:46] But there's a catch. [01:47] These tools only work well when they have deep access to company systems. Your copilot needs to see your entire code base. Your chat bot needs to 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. [02:13] 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 up market faster. Think of it like Stripe for enterprise features. [02:43] day. [02:44] Zach, I'm so excited to have you here because I feel [02:49] Like I maybe turned you. [02:51] into an AI fiend at this point. Before the show, we were talking about how many tools you're now using. So before we dive in, can you just tell us a quick list, maybe not so quick list,

[03:03] of all the AI tools the technology team at LaunchDarkly are now using. [03:08] Yeah, absolutely. Let's see. So on the design side, we're again, we're exploring a bunch of things. So there's going to be a bunch of tools. So lovable. [03:18] V0, Figma make, we're using on the product side, obviously, we're using chat PRD. And then on the engineering side for code, [03:28] heavy cursor users, heavy Devon users. We're also using now like cursor's background agent. [03:33] I personally use windsurf because I like windsurf, but most of the rest of the work does not. We are using we're trying augment. We are looking into cloud code. We're doing all the things we're also looking into PR review. So we use Copilot for code review and we use cursor for code review as well. [03:51] Okay, and I feel like 18 months ago we were using... [03:56] Maybe a little GitHub copilot, but not much. Yeah, not much more. And one of the things that I really liked that you and I did together and you were a champ in coming along this journey is... [04:07] we really decided that [04:08] in order for AI to be really effectively adopted by a team like LaunchDarkly's engineering organization, which is over 100 people, we really needed to put some concerted effort behind it and put a person in charge. And lucky you, you drew the either short or long straw, however that works. You know, what do you think about teams approaching this kind of engineering-wide transformation? And what kind of organizational and cultural things you need to do to make it possible?

[04:37] I do think having a person who's kind of, I don't know if in charge is the right word, but whose responsibility it is. [04:44] to drive that kind of change. And I think that having someone... [04:49] who's close to the code helps a lot because you don't really know [04:53] what's working and what's not working unless you're in the code, at least on some basis. And so that can be a manager, that can be a director or someone, but it has to be someone who's actually trying these things, I think matters a lot. And yeah, you, I think you were looking for someone to take that role. And I was [05:10] I think skeptical, right, of how well things were working really well for you over on your side job on chat PRD. But when I tried to do the same things in our code base, I was struggling. [05:23] And so I really came at it from a standpoint of, [05:26] I want to understand what works and what doesn't and and [05:30] either be able to push back on you and say, hey, you know, like, it's great over there. But on this, you know, larger code base, it's not working, or be able to actually drive change. And now I'm at the like, [05:42] I'm on board. Let's let's drive change. I think that matters a lot. [05:46] Yeah. And one of the things that I think is really important for our listeners, especially ones that are at growth stage or larger companies, is. [05:53] Vibe coding is not an acceptable enterprise development strategy. Like I love it. Right. I can do 100 commits a week by myself on my side project on my startup. And, you know, I can recover from quality issues. You know, the maintainability of the code, the issue right now, that's not my big issue.

[06:13] business issue. But when you're working on a code base and a platform like LaunchDarkly's that powers trillions and trillions of experiences every day, [06:23] You can't take the same strategies and tactics that a vibe coder could take and bring those into not just an individual developer's workflow, but an entire team's workflow. And so what have you kind of discovered as you try to figure out how to make these tools work for developers? [06:41] a larger team. [06:42] With smaller teams, you have more flexibility in terms of how you approach these things. With larger teams, you have more enablement and stuff like that. We're kind of in the messy middle. [06:52] I found it more difficult to [06:55] sort of like operationalize that and make everyone successful. And what I found was [07:01] Everyone was on their own journey. [07:03] to try to be successful with AI [07:06] And that just doesn't scale very well. Right. So, yeah, you really need to come up with a system in order to [07:13] to make everyone more successful, right? What I want is when those skeptical engineers [07:18] jump in and they try cursor, they try whatever for the first time or they try it for the first time in a while. [07:24] I want them to be successful so that they get that aha moment and, you know, [07:29] If you just leave them on their own, then you're not going to get there. [07:32] Totally. And we had, I think you and I had mutually had the experience of developers having their first experience actually be negative, whether it was a cursor or with Devon. It was like, see, I knew this was never going to work. And here's my first pass proof that it didn't work. And so you really did a lot of what I appreciate about you is you did a lot of technical work.

[07:52] to make sure people were successful. We'll definitely talk a little bit more about some of the kind of culture and operations piece, but I actually want to get dive into what you did in the code base to make it easier to work with Cursor and Devon. So can you walk us through some of those things that you did? [08:07] Yeah, absolutely. So. [08:10] In my IDE here, you can see one of the things that I realized is [08:14] what's good for [08:16] humans is also good for LLMs. And so I really started with how do we make sure [08:20] that the repo is well set up for humans to know how to work in it. And so we have this docs directory. And I pulled a bunch of stuff from Confluence, from Google Docs, from other places in the repo. [08:31] And I put it all in here, right? So we have front-end organization. We have accessibility. We have a JS style guide. So it's this very detailed documentation that we've put [08:42] into the repo itself rather than have it in other places. And this way, LLMs can access it, humans can access it, et cetera. [08:49] Okay. [08:50] In addition, we had these, we had cursor rules before, we had a CloudMD file, and I wanted to consolidate that. And so instead of a .cursor rules, I have this .agents rules, and the idea is to [09:02] kind of centralize all of this knowledge in one place. And so you can see here, [09:06] I have something like TypeScript Essentials, which has a really... [09:10] kind of like quick... [09:12] like the quick hits of what's really important. And then it also links off to like the comprehensive docs and says, hey, go, if you want to find out more, you know, go look at the JS style guide.

[09:24] And so then our cursor rules [09:27] actually just point to that. Right. So with our cursor rules say, hey, if you want TypeScript guidelines, go find this file in dot agents and then [09:36] I talked about augment earlier. We were trying augment. I set this up yesterday. [09:42] And I had, I asked the augment, [09:44] agent to just create this. I pointed at the cursor rules and I pointed it at our agents rules and I said, "Can you just create this file?" And so it [09:55] It did the same thing in this way. [09:56] We don't have to duplicate everything across multiple tools or tool files, and it's much easier to get stuff working well by default. And the whole idea with this is, again, I want people to be successful out of the gate and having this kind of centralized place, having all this documentation in the repo. [10:14] just makes it way easier for tools to be successful by default. [10:18] Yeah, one thing that I hear a lot is people are really frustrated with the tool specific rules. They're like, why do I have a Claude.md? Why do I have a cursor rules? Why do I have these GitHub rules? Especially if you're experimenting with the number of tools that you're trying. Each tool has isolated their rule set. [10:36] in an individual file structure. And I think all the engineers out there are like crossing their fingers and hoping that there's like one rules protocol to rule them all that shows up. And I think what you've shown is you can just create that yourself, create a directory in your repo, [10:52] put a consolidated set of rules, make your sub rules for each of those tools point to those rules. So say when you're looking, you know, when you're working on front end, reference these rules. And then that makes it much more scalable. The other thing that I want to call out for you is

[11:08] I think what you said at the beginning, you're using, I don't know, a dozen tools, probably like three to five IDEs across the engineering organization, probably within any individual engineer. [11:20] You're testing like I want cursor today and Devon tomorrow. And if you don't have your rules set up for all those tools, then you're starting from scratch every time. And so. [11:31] I really like this idea of a rule set up and then consolidation. I'm curious, do you feel like the rules have improved the quality of the outputs significantly? Yeah. [11:41] Yeah, absolutely. So here I'm looking at our feature flagging rules. And it's interesting because we are a feature flagging. We have a lot of feature flagging code in the code base. And one thing we noticed was [11:52] that some of the models, some of the tools would get confused about whether we were asking it to [11:57] create feature flags on the like launch darkly product or whether we're actually trying to get it to do stuff in the code and so there's there's a bunch of stuff that i did to just be really specific about how to make how to be successful when creating feature flags i want you to return a link that kind of stuff [12:13] And it really has made a difference. Yesterday, literally, one of our product managers was doing a task with Devin and was able to tell it to put the feature behind a flag. And Devin went and used the MCP and hit the flag and everything worked. [12:33] So I have another question about rules because LaunchDarkly's giant monorepo is

[12:40] 10 years old or something like that. It's got a lot of code in it, front end, back end, tests, all this stuff. What do you think, if you had to give some advice to peer engineering leaders who are approaching the same problem, what are the must have rules from your point of view in the code? You know, I saw like a lot of front end stuff, but what, you know, what are the quick hits of what you think should belong in a kind of cursor rules or a rule set? [13:01] I would say the best tip is... [13:04] ask the agents right to get you started. And so Devon actually has a great wiki that for each repo that Devon works on, it creates a wiki and it has a ton of really good information in it. [13:15] And so I actually started with Devin and I said, hey, can you like this is what I'm trying to do. Can you create. [13:21] basically the human-readable docs. [13:23] for this. And so Devin did a pass and created a bunch of docs, suggested some structure. We went back and forth and I kind of tweaked things and then I took that output [13:33] And I went through it with kind of like a fine tooth comb, because I think it matters, right? It matters to get those details. [13:39] write. And then once I had the human readable docs, I went to cursor and I said, hey, can you take these docs and can you take your existing cursor rules and can you turn those into a dot agents file? And so it was a combination where [13:54] you can kind of lean on the agents a little bit to [13:58] help you get unstuck and get started, and then also use your knowledge of what's important in the repo. [14:03] And the other thing is that I was looking at where are people getting stuck? I knew that people... [14:09] on the front end would struggle with

[14:12] Getting like testing basically writing unit tests. It would write a [14:17] Jess tests and we use V test and stuff like that and so putting in specific rules to make sure that [14:23] where people get stuck [14:24] we have rules to help the agents be more successful. Would you mind pulling up [14:30] Devon and actually giving an example of generating rule and I have an idea for you. [14:36] Which is like rules around generating data visualizations since we've done so much and just, you know, see what it comes up with. [14:41] Yeah. JOHN MUELLER: Here's an example of both asking Devon Wiki a question and then also using Devon to create a rule. So we can say-- [14:51] to the Devon Wiki, we can say, what are the libraries used for charting on the front end? I'm curious, while this is loading, how long did it take you to set up [15:03] Devon's environment. It's something that, you know, everything's easier with a little VibeCoded Greenfield app, except for setting up Devon's environment. It's just as hard. I'm curious what your experience has been configuring Devon to work in a large repo. [15:17] Yeah, I would say to get up and running with Devin, I got started pretty quickly and [15:24] We have kind of, we have a separate flow for front end and back end. We have a concept of sort of like front end only mode, which proxies against another back end. And so I was able to get Devin's machine up and running pretty quickly in just front end only mode. And then I was able to take on front end tasks using Devin. One of our other engineering managers actually came in and saved the day on the back end to get the full machine.

[15:49] like running our whole end to end, up and running with Devin. And that took him, I think, a little bit more time than it took me. But the nice thing is you can do that sort of incremental, you know, you do what works, you don't have to have Devin running your full app locally in order to get value out of it. And so it's just about kind of like doing it piece by piece. And again, if it's hard to get Devin, [16:12] Up and running, it's probably hard for your human developers to get up and running, so there's always [16:16] incentive to make those things better. [16:19] Yeah, I will give just because you said it, you said, you know, Devon's environment doesn't have to be running for you to get valued. My number one Devon prompting trick is don't run this locally. Just give me the code and I'll test it for you. So sometimes I bypass that process entirely. OK, so you asked this question of Devon Wiki. What do libraries use for charting on the front end? And it gave answer recharts. [16:42] some other things. And so what would you, one, is this information pretty accurate? And two, what would you do with it? [16:48] Yes, it is accurate. We are using multiple libraries. And so that was one of the things I was curious about is we've brought in [16:55] several libraries and we're kind of trying to figure out how to consolidate and so it picked out that we're using recharts, we're using VizX. It lists eCharts as a secondary library, I don't know if that's strictly true, but generally this seems very [17:08] Correct. And so I like to use Dev and Wiki to just sort of ask basic questions about the repo, make sure I understand [17:15] what we're doing. But if I actually wanted to create a rule so you can't take action from from Devon Wiki. So I what I want to do is I want to create

[17:25] a new document, [17:27] a human readable, human centered document in our doc/frontend, about how to use charting libraries. And then I want to also add, [17:36] a rule to dot agent slash rules. So I'm just going to give this to Devin, and I am going to see how it goes. So Devin's going to spin up a new-- [17:45] session here. [17:46] And one thing I want to call out for folks listening on the prompt is you specifically said you wanted to create a markdown document markdown is every engineering agents favorite file type so that's a good way just to give a little bit of structure to your code it also tends to. [18:04] pretty print and be human readable and easy to view in GitHub and all that stuff. [18:10] And so what you're doing here is just asking Devin to make those docs for you. And this is one of my favorite use cases of Devin. I think you know this. It's my favorite Devin hack, which is I have a GitHub action on every PR that is [18:27] writes docs for the PR and adds to a change log programmatically with Devon. I found that it's a very good technical writer. [18:35] You know, sometimes the code is okay, but the technical writing is very clear and very good. Yeah, I think that's exactly right. The Devon wiki is very good. It knows a lot about your code base. It has this very explicit... [18:47] way of learning and understanding your code base. And so it is very good about kind of describing that back and as you said, doing it in a solid technical writing way.

[18:58] Yeah. And then one of the other things that I want to call out for people that are maybe listening and not watching is, [19:02] We are chit-chatting because we're waiting for Devon to spin up a virtual machine. So for those that don't understand kind of how Devon works, it actually spins up a virtual environment that reflects a development environment. It's going to open it up. It's going to read your code base. It's going to do all this stuff. And so, you know, it takes a minute to actually boot into an environment a little different than running something like Cursor locally. [19:25] Yeah, that's exactly right, where these other tools are. [19:30] Just using whatever you have locally, Devon is running its own machine, which has a lot of upside. It can run a browser and see a browser. It can do a lot of things that don't come out of the box with these other tools with the downside that it takes a little time. Depending on your repo, it takes a little bit of time to actually set that machine up and get it running. [20:00] line incentives on getting your repo to work well for both your agent co-workers as well as your human your human colleagues that is that is my favorite thing is all the things that have been hard for humans forever and we have just kind of swallowed it and said well that's the way this works become even more important uh today with with these llm tools to to solve and improve well

[20:30] to solve and improve them. If I said, you know, two or three or four years ago, Zach, go document everything. [20:36] in the repo high quality human readable docs you just go do it by yourself it would take [20:42] forever to generate high quality docs that, you know, really reference our code and understand the nits and details and [20:50] I think the fact that even you can spin up docs so quickly is so transformational to how you can, and I know we'll see this in a little bit, like burn down tech debt, make your engineers happier. And so I do think, you know, a lot of engineers have the skepticism that adoption of AI tools is really about moving faster, shoving more junk in the code, like just getting feature bloat. [21:20] to take care of some of the things that you have just hated forever in either how you run your software, how your team operates or the code itself. And so that's one of the advantages I think people underestimate in larger organizations because they blur the line in their mind of AI assisted engineering with vibe coding, which is not what we're talking about right now. [21:41] No, not at all. And technical debt is my favorite use case for AI to supercharge medium sized organization. Okay, so what we're seeing here is Devon's looking through your repository, accessing its knowledge. Actually, I'll take a pause here. [21:58] Have you set up the knowledge in Devon explicitly, which is for folks that don't know little snippets of kind of rule? It's almost like Devon's rules in some ways, little snippets of knowledge and rules. Have you set those explicitly or have you simply accepted and approved the ones that Devon suggests? Yeah. So as as you mentioned, one of the things that I really liked about Devon, especially as we were first getting started, is that.

[22:24] it builds up this knowledge on its own in some ways. So as you're interacting with Devon, it will make suggestions for additions to its knowledge so that it gets sort of like, quote unquote, smarter every time. Where some of these other tools have memories, because Devon's starting a fresh session every time across different users, [22:43] it has a centralized knowledge kind of like repository and so [22:49] It's been a mix. We've sort of let it build up over time and various people have accepted knowledge. I added some knowledge very early on. I will intentionally add stuff when I run into problems. But then again, when I moved all this documentation to the repo, [23:05] and I was trying to centralize everything, Devin's knowledge now primarily points to that same .agents directory, because I don't want to have the duplication. I want it to work for all the tools. I don't want just Devin to be effective. I want all tools to be effective. [23:21] Got it. So you've really taken all your tools, whether locally hosted, [23:25] in the repo or cloud hosted like Devin and just made this agents folder. [23:31] the source of truth. [23:32] That is exactly right. [23:34] How I AI is now on Lenny's list with my personal selection of the best AI engineering courses on Maven. You can spend months thinking and playing with AI before really integrating it into your workflow or shipping an actual AI feature. [23:51] If you want to start building, then these hands-on Maven courses are for you. Learn directly from Aishwarya Naresh Riganti, MIT instructor and AI scientist at AWS, or Sander Shuloff, who has authored research with OpenAI, HuggingFace, and Stanford.

[24:10] To pivot into an AI role or successfully lead your company's next AI initiative, visit maven.com slash Lenny to enroll now. Use code Lenny's list for $100 off. That's maven.com slash Lenny to get ahead in the AI era and start building. [24:33] So Devin has a plan now. One of the things I like about Devin is it gives this confidence now, like how confident is it in the task at hand, which is... [24:43] is nice because sometimes it's not confident and you it's better not to proceed. This is something that, as we mentioned, Devin should be really good at. And so I feel good about [24:55] its ability to execute this, but it will give you sort of an overview. If I thought, if I read through this and I didn't like what it was doing, it's going to run prettier on the markdown files, which actually I think is a good idea. But if I didn't think that was a good idea, I can update its plan while it's deciding what to do next. [25:10] Yeah, the other thing that I enjoy about Devin is nine times out of 10, it's confidence gets higher as it goes. So it always starts like medium confidence, but I have to investigate and then it's like high confidence, I know what to do. [25:25] But occasionally it fails me deeply and I have bullied it so much that it starts to progressively lose confidence and then it's like low confidence. I haven't been successful so far. So I find the confidence assessment pretty accurate.

[25:41] Yeah. [25:42] Okay, so now it is creating, it's created multiple Markdown files. So it's created a charting libraries.md file. [25:50] And we can actually, if we want, we can jump over. So there's a shell. There's also the code. So I can actually go look at what it's creating. [25:57] while it's creating it. [25:59] Um, [26:00] So charting libraries guideline, it's creating that in our dot agent slash rules slash frontend. [26:06] It looks like I think it also created one in docs. So this is the human readable version, which I'm not going to go through in detail, but looks [26:14] It has examples, has the different libraries. I like all of that. [26:19] and then [26:21] In the agent's rules, it's sort of a consolidated, you know, must use this when. I like seeing that. I would go through here and really make sure it's accurate in what we want. And then... [26:32] It's a little long, I think. For me, I want to keep the... [26:36] The agent's rules pretty concise. So you're not leaving the context and just so it's not too much. And I would also want to make sure that it links out to the full documentation is another trick that I like to do so that a tool can decide to pull in that additional context if it wants. [26:52] Well, one little trick that I learned from another How I AI guest is that if you notice, Cursor reads long files in chunks of 200 lines. And so his goal was to keep [27:05] these files under 200 lines so that it's not chunking the content. And so I saw yours is just like a little bit over 200. So one of the things you might add to your

[27:16] rules for rules is try to keep your rules files under 200 lines, for example. Now, again, I don't, I don't know if that's actually full or true. But it is a tip somebody gave me so I'm passing it along [27:28] With no personal context. No, I mean, that's actually, again, that's a good tip for humans, just like it's a good tip for LOMs. And you said something that I think is really interesting, which is I actually I have a read me. [27:41] I have a human read me about the rules so that people understand how to create new rules, but I should actually probably have [27:48] something geared towards LLMs so that when LMs are adding new rules, they're doing a better job of it. [27:54] Yeah. Okay, so I see this. It looks great. So it's created a human readable docs in your docs folder, a rules for your LLMs. You're gonna review this, you're gonna do a PR and merge those docs into the repo, maybe take a look at them, edit them. And you've used, you know, Devon Wiki. [28:16] Devin agent... [28:18] And then it's spun up this code base to write those docs. And so I think this is a really great flow. I think people are going to learn a lot from this. You know, one of the things you said earlier was that tech debt is your favorite use case for these AI tools, I love. [28:34] love to hear it because this is how I try to pitch senior engineers and senior engineering leaders like you to really adopt these tools when they're really skeptical. Can you walk us through how you actually approach

[28:46] burning down tech debt using these tools where it's made it easier maybe. [28:51] Yeah, absolutely. So here we are in cursor and I'm going to show you that same [28:57] Agents directory there's so I showed you agents rules before we also have agents migrations and so this has a couple files in it. It has a CSS module conversion file, which I created to help us. [29:11] Convert CSS files to modules CSS modules and then it also has I just added this one the other day, which is the one that I would like to show and so. [29:21] What it is is basically it's a combination of instructions for agents and a checklist basically like a task list of what to burn down. And so [29:33] The problem that I was running into is that our front end [29:38] unit tests, when you run yarn test, it just there's so much noise in the console that it's really distracting. There's some actual legitimate problems in there that are just kind of being warned about and ignored. And I wanted to pay that down. But it's [29:54] It's one of those things that is annoying, [29:57] but it's not quite annoying enough for someone to own it. And also it's such a big problem, it's really hard for one person to just kind of like [30:04] take that and pay that down. And so well, and I'll say, imagine as an engineer, you go to your product counterpart, and you're like, hey, [30:14] I just want to spend like a week or two just making our test logs just a little less noisy. So my life's just a tiny bit. Like it's such a hard pitch to make for...

[30:25] work like this it's super important and like the pitch can work on the right leader but again like this is the kind of thing that's hard to justify in a fast-moving org [30:33] Yeah. So I'm actually, I think what I'm going to do is I'm going to ask, I will talk, I will talk through kind of how this works. And in the background, I will have cursor take the next stab. So I'm actually going to say, so it has this context. It knows I'm in this file. And I'm just going to say, can you take the next tier of tasks? I can see here, there's a tier one, tier two, there's three files. I think that's reasonable and fix them. And I'm just going to, [30:59] So you click Go. [31:01] and we're going to see what happens. Okay. So in the meantime, what I did to actually produce this is I ran yarn tests and I piped the output to a log file, which I'm not like a super techie tech person. And so I actually asked cursor how to do that effectively. And then I had a log file and I gave that [31:21] to Cloud Code. [31:24] and I asked Claude to basically create this file, [31:28] And so what it did is it went through all-- it actually had trouble with how big that file was, but it was smart about working around that. And so it found out that we have something like 1,200 [31:39] extra lines in a test run that we don't need to be there, that we don't really want there. And then it quantified this or it sort of grouped this into [31:49] different types of warnings. [31:51] And then [31:52] which files are the worst offenders?

[31:55] And so then once we had this file, I said, great, can you go fix the tier one worst offenders? And so it actually went and has done that successfully. That's been merged in, reviewed and merged in. [32:08] And then [32:10] I can do stuff like this where I just say, like, for any-- the thing that I like about this is you can just give this to any agent now. I can slack Devin and say, @Devin, can you pick up the next task in the front end test noise cleanup? I can do it here in cursor and watch it go. I could give it to cursor background agent. [32:29] It sort of like makes it easy to pick these things up as... [32:34] as individual tasks and make progress on them. What I like about this approach as well is it's very there's a lot of parallels to how you would approach something like this with an engineering team of human partners. Right. You're going to take a problem. [32:50] somebody's going to go investigate it and identify priority tasks to do. You're going to put those tasks in some sort of task tracking system and [33:01] You and I both know all of our beloved task tracking and project management systems. And I am starting to see. [33:08] cursor markdown files become the new task tracking system. So I'm seeing this trend of these checkmarked files in cursor just being the source of truth for progress on initiatives. So you created basically a list of epics and tasks here, if that's what we call it, and they're prioritized by how severe they are.

[33:38] And then what I'm presuming you're doing is the work happens. [33:42] whatever agent you decide to do the next task, [33:45] closes it out. [33:46] You review the PR. You make sure any changes work. [33:49] You merge it, it gets marked off. The other thing I want to call out is while you are probably running this yourself, [33:56] You could probably also... [33:58] get more people on the team to be aware that this test exists and just say, hey, if you have a few minutes and you're able to review a next set of noisy tests, like, [34:06] Tell Devin to pluck off one and do the code review for me. And it's all set up and it's ready to go. So, you know, I think this multiplayer aspect is very important when how you approach, you [34:18] some of these tools when you're working in a larger, larger team. [34:21] Yeah, just today I had a PR app to fix a few stray errors on this one file. And one of the people from the team that works primarily on that, I included them in the review, [34:35] And he said, hey, if there's any more... [34:37] uh, stuff like this, feel free to kind of like throw it over the wall. [34:40] to us, you don't have to be the one that does all of this. He didn't know that I was just using Kurt or Claude or whoever. And so now I can actually just point him at this file. I said, hey, take a look at this file. And if there's any ones you want to pick off in your ownership area, [34:55] then just go ahead. And you're exactly right. Doing that, democratizing that, this is [35:02] Great for, again, I'm saying the same things, but it's great for bots. It's also great for humans. Humans can come in here and understand this and...

[35:10] work against it. Yeah. And if you're feeling like, you know, crafting your farm to table code and you want to pluck one of these off yourself and you want to fix it, you can approach it the same way, right? Just open the file, mark the thing as done, do the PR. And so I really do think it's important that folks think of [35:29] kind of these tools as an extension of the team. And the more the tools can operate the way the team would operate, and the more the team can operate in the same way the tools can operate, then we can kind of all collaborate together and be much more efficient. So I think this is a great [35:46] Super great example. I'm not going to make all of us watch Cursor go through tests and lint errors because I have [35:55] lost enough of my life to doing that. That is a really, really great example of [36:00] Tech debt and then just to ask the question, [36:03] What's the end payoff for [36:06] front end developers like [36:07] the actual issues bubble up in your tests and these tests get less noisy? [36:12] Yeah, I think one is it's easier to find stuff when stuff is going wrong. Two is, I think it said that the biggest problem was actually accessibility warnings. So that's like a real problem that exists. But when there's 1200 lines of that, and a lot of that's coming from like the same component, if it's tested a bunch of times, will... [36:31] will spam the logs. But being able to sort of surface the actual signal through the noise, I think is one of the key benefits. [36:40] Okay, and then for our last workflow, I know that...

[36:45] Zach, you're going to impress everybody and everybody's going to think you are just an AI enabled, you know, cutting edge engineering leader who only works with his army of bot friends. But yeah. [36:57] You're actually hiring at LaunchDarkly. We expand the team. You're always bringing in great talent. And you've actually used AI to solve another problem, which is making sure that you're doing a great job hiring. So do you mind spending a couple minutes on what that little workflow looks like? [37:27] with. It's not a direct report. I don't love just dropping and being like, Hey, this isn't great. But we were trying to improve, make our hiring more consistent. And I [37:38] I created a rubric for all of the panels that we have. So there was really clear guidelines about how to score a candidate. But the other piece of it was [37:48] we needed people to follow those guidelines, and I wanted to be able to give people feedback about, [37:53] whether basically I wanted to raise the bar of the actual scorecards that we were creating. So I created this custom GPT, I gave it the rubrics and I gave it examples of [38:03] good scorecards, bad scorecards, and give it as much kind of like, you helped me write the prompt. So thank you very much for that. And so what I'm going to do is I'm actually just going to paste in a scorecard. And so this is, you know,

[38:19] a scorecard that we got, and I'm going to click go. [38:23] And it's going to do a few things. One, the rating that it's giving me is the rating of the scorecard itself. So it's a little meta. But I want to know, basically, one of the things I did in the prompt is I said, give it a rating. Is it excellent, good, fair, or poor rating? [38:42] in terms of scorecard. And then I want you to list out strengths and potential improvements. And so, and then the last thing that I had to do, which I also think you helped me with, so thank you very much, is what I, the format I wanted to give this to people in was [38:58] to send it to them over Slack, right? Like, hey, like, thanks for doing this. But also I had a little bit of feedback, and so [39:05] Thank you. [39:06] It actually gives me the detailed feedback, but then it also crafts a short Slack message that I can, if I want, just copy and paste and send to the person who created the scorecard. [39:17] I love this because so many managers and hiring managers can empathize with this, because if you're running an interview panel, you're having. [39:25] everyone from your boss to your direct reports to people you've never really worked with directly interview candidates. Right. So you have these cross-functional interviews. And while you can have all the rubrics in the world, [39:37] interviewers, [39:38] sometimes write terrible notes and, you know, assess the wrong things or don't give you the right details or really using the rubric incorrectly. And you're not sitting in every single one of those interviews to give live coaching. And so this is a really nice way to

[39:55] the kind of standards very high, and then giving you some leverage as a manager to give your team coaching. And then as they get this coaching, they get better at doing the interview feedback, and then you can be more confident in your hiring decisions. [40:09] Yeah. And honestly, this helped me as well. In order to test this out, I was doing a bunch of interviewing and I was writing scorecards and I would pace it in and see what kind of feedback it gave me. And it was giving me very good feedback. And I learned very quickly the kinds of things be, you know, be more specific, you know, avoid avoid certain kinds of things. And so it actually made me. [40:30] write better scorecards just through trying to create this tool for other people. [40:35] Okay, Zach, you have given a masterclass in how engineering leaders at larger companies can really approach integrating AI into not just their individual workflows, but their team workflows. So just to cover what we talked about. One, [40:49] let the team experiment. Like every tool, just let's see what works. So you seem pretty kind of generous with your experimentation mindset around what tools can use, can bring value to the team. I think two, context is king. And so you're loading up your actual repository with docs and rules. Three, those rules are centralized. So you don't use agent or tool specific rules. You create a central agent repo, and then point all your specific

[41:19] Those rules you [41:22] use Cursor and other tools to create plans to burn down tech debt and then have those AI tools burn down that tech debt. And then since you have all this free time now, you're coaching yourself and your team to be better. [41:37] interviewers and better hirers. So just that. No big deal. That's all you have to do. A few things, yeah. And this is all, I mean, truly, from a personal professional development perspective, these skills were developed, what, in the last 12 months, right? Right. [41:50] Oh, I think January is when I really started, like, took on the mantle and playing with Devin and really going down this path. So six months and we didn't give you, I mean, I think we offered, but like formal L&D, none of that. We just pushed you into it and said go. [42:07] Yeah. [42:08] Okay, I'm going to wrap with two quick lightning round questions and I'll get you back to all your AI assisted code. Question number one, you listed so many AI tools. [42:18] - Which one? [42:19] is your favorite or which one has been most transformational [42:22] Ooh, that's really hard. I would say windsurf, actually. Everyone was... [42:28] hot on cursor and I, it just wasn't the UX at the time was not clicking for me. And I saw a video for windsurf and I was just like, whatever, I'll give it a try. And I had the free trial and within an hour, I think I was paying for it because I just, it really clicked for me and the agent workflow just really quick clicked and I was hooked. [42:49] Amazing. And then when AI is not listening to you, you're such a, you're conflict avoidant. So I'm actually very interested in your answer here. AI is not listening to you.

[42:59] You need to give it harsh feedback. What are your tactics? I know you don't yell. I know you're very polite, but what do you do? [43:07] I mean, sometimes I lose it, but I have to I [43:11] The thing that I actually do is [43:15] sometimes I just feel like it's not the right task. Right. So it depends if I if I think it's something that [43:21] AI should be good at [43:22] then I get a little snippy with it. Maybe I don't yell, but I'm definitely, you know, getting getting a little annoyed. But I also think that sometimes it's okay, right? Like sometimes it's not going to work and you don't have to keep banging your head against it. And I think developing that sense of where it works and where it doesn't, it's been really powerful for me. And also, sometimes I just like getting in there and getting getting dirty, getting getting my hands in the code. And so, [43:48] Yeah, I think my technique is actually... [43:53] Either I do it myself or [43:55] I go back and try and fix it. You know, am I providing the right context? You know, what, what, what is missing that it can't accomplish this effectively? [44:04] Yeah, you're a very good manager. So I think it's from those skills. All right, Zach, this has been super informative. [44:10] Where can we find you? And is there anything we can do to be helpful? I'm on LinkedIn. We are hiring at LaunchDarkly. And also if you are a LaunchDarkly user and you have any feedback, I love user feedback. So please send it my way. Amazing. Well, thank you so much, Zach. [44:26] Thank you. 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.

[44:35] 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 howiaipod.com. See you next time.