Exploring with agents (Interview)

26:44Python, I think they're called Python enhancement proposals. And I just swapped out Python for projects. So they're called project enhancement proposals. And that's where the work gets done. So if it's a specification, it could be some research. It can be a knowledge-based article. It could be I even have my agents write blogs. I call them builder logs. It's so cool. When we're done with a journey, I'm like, hey, can you just write a story about that? And I use it to learn because I'm often exploring places where I don't know a lot and I need to get caught up. and I'm not the kind of developer who just wants to sit there and say, just do all the things, prompt a magical way and don't have any understanding.

27:23A lot of the reasons why I'm building software isn't just to build the software and solve the problem. It's to understand the software that should be written to get there. And maybe in the future that doesn't matter. But all I have to say is that I think a lot about this process of getting to good software. And I question, you know, is my software worthy of open source? Does somebody else want to use my stuff? And how will open source change when essentially generating a solution, whether you have taste or not, is the closest version of free we've ever had.

27:56Amelia Wattenberger:Yeah. One, I want to check that out. That sounds really cool. Also, you're really good at naming things. Asian flow is cool. Two great gems in there. And two, just like, what does the world look like when we have software on demand? than, I don't know, software too cheap to meter. I think whenever I think down the path of like, oh, what if software is just instantly generated on the fly? I always come back to like, people don't want to make all of their software, right? Like, I don't want like a blank phone and to be like, oh, you know what, there's folders and I have an app where there's like emails in here, right?

28:41Amelia Wattenberger:Like, I don't want to have to think about that. I want like these smart defaults or like, like tasteful things that I've decided that this is the way I want to work because I see it out there in like its fully fledged form, right? Like doing a lot of abstract thinking and product projects, talking about abstract things is really hard because you can say words and there's a picture in your head. And then the person you said those words to has a completely different picture in their head. So like the farther you are away from something tangible and something fully fledged, the harder it is to communicate about, even if LLMs are, you know, so good at English and language.

29:23Amelia Wattenberger:um so like there's got to be some room for taste and like opinions and um like having proposals for the ways things are put together and how software works in in that world and like maybe things look more like some kind of like remix culture right where i i say like i want to use atoms and then like i can change it right like i can change it on the fly but like i i don't want to go from like zero to one all by myself. I want to like go from 0.5 to one. I'm going to go from 0.5 to one. I've thrown a lot of numbers around and I don't know if they all make sense. Well, you know, zero to one is a well-known term.

30:06I like zero to 0.5. That's because that's where I feel I'm at.

30:10Amelia Wattenberger:I feel like a lot of my things are at 0.5 and 0.5 is usable by Adam and Adam likes it. Okay. I like my stuff. I'm constantly working on it. Hey friends, I'm here with Dan Mangus, co-founder and CEO of RWX. Dan, what makes RWX and the way you're doing CI so different and interesting to our audience? Obviously, we're talking to you because we want to promote what we're doing. We want more engineers to become aware of what we're doing at RWX. But I think the thing that's interesting to me is that RWX is really kind of the first major evolution in CI and the approach for CI. And this is just highly relevant with agentic-driven coding.

30:50You know, CI has largely been the same since the advent of the practice. But these platforms were created when being able to run code in the cloud was really valuable. The fact that you could spin up virtual machines that would run some automation on a Git push was, you know, really impactful for engineering teams trying to, like, build good developer processes and tools. But that's kind of the extent. What we've done at RWX is we've taken state-of-the-art techniques techniques used in build systems at organizations like Google and Meta. Google has their internal build system Blaze, inspired the open source Bazel tool.

31:25But every engineering team I've talked to that wants to adopt Bazel has just found it extraordinarily difficult to use and configure. You have to have a dedicated engineering team to build and maintain the rules. It's hard to extend it to work with different types of languages and frameworks that engineering teams are looking to adopt. So it's been too prohibitive to actually adopt those technologies. But the ideas behind Bazel are really impactful. They're similar to a lot of the ideas behind Nix. I would say Nix is kind of very similar in the difficulty to adopt. And effectively what we've done at RWX is we've taken those techniques, and we've made it very easy for engineers or agents to actually adopt and utilize those, which namely are the automatic content-based caching and the graph-based task execution, execution, which means that RWX eliminates all redundancy.

32:12You know, whereas other platforms are having to run the same setup steps on the same jobs and every virtual machine that's spinning up, RWX can run the setup once on one machine and then fan out accordingly based on just your dependency graph. So effectively with RWX, you never have to think about parallelization at all. You know, on other platforms, it's always like, well, do I add this onto the existing job? Do I make a new job for it? But then I have to duplicate all that setup. With RWX, you just define the tasks that you want to run and the dependencies between it, and we will run it with maximum parallelization based on your dependency graph.

32:45Well, friends, a good next step is to go to rwx.com, learn more, check out CI in a whole new way. Once again, rwx.com.

32:59I was telling you before that I feel like I get so far into one thing I'm trying to get it to be usable by the masses. But to get there, I've got to solve some other problems. So one thing I did recently was I created this thing called Gauntlet. I just bought the domain and it's going to send my software into the Gauntlet. So rather than CI, it's really like a battle tested ground. There's ephemeral Incas containers involved. There's, which is, do you know what Incas is, Amelia?

33:28Amelia Wattenberger:I do not. This is the coolest new thing, by the way. It's not new. It's Canonicals LXD, re-licensed because they had a license challenge. And one of the original creators, maintainers is a part of that. Forked it happily. Stefan, I forget his last name in the moment, but Incas essentially is system level containers and virtual machines. So think of like a hypervisor where it can instantiate an LXC container or a VM pretty easily on something like Proxmox. something like that but super super fast and built on top of zfs and just phenomenal piece of software and so here i am trying to solve this one problem and then i'm like i get so shiny object right shiny objected and i i have this idea to like turn my code base into a sandbox and i have virtual machine over there on proxmox i'm like listen agent let's just build out a sandbox and prod is prod and sandbox is sandbox and maybe the developers who have been down this road this is like normal for me i'm like okay this is a brand new world let's figure out what sandbox feels like we can break everything in here we can tear it down and build it back up again in fact that's what we want to do because if we can that means prod is ephemeral in a way and we can get back to zero or get back to one quickly and so as i'm building this thing out and doing all this stuff i'm like there's a product here adam because like i want to do this not for just this software but for other things i'm working on and so i started to get into this point where like i stopped working on the main thing you know and now i'm over here working on this side project this side quest that actually is really really cool so that's it like that's what's happening i was like i'm trying to go this direction but i get shiny objected and now i'm over here building a side quest that's kind of cool are you doing things like that too or is it is it just like you just go straight and you're good there's no straight in my world there's no straight gauntlet sounds very cool there never has been and it's even worse now than than it has been right like it's so easy to build things that we're we're like doing it more and more so so we wanted to talk about uh intent which is the app that i've been working with with augment code um and a lot of the motivation for that is like we there are existing problems within software development that like we have never solved and we're kind of just accelerating things.

35:50Amelia Wattenberger:And it's kind of like, can we just solve these first so we can accelerate faster? So there's actually a project I did at GitHub Next called like Workspaces or something where GitHub's fully remote. And this was right after COVID. And like to work collaboratively with other developers has never been easy, right? Like GitHub says it's like the social hub for developers, or at least they used to. But we don't really work together before the PR. Before that, we're kind of just in our own heads. Is that right? PR is where it happens. In my experience. Yeah, I guess the old way. The new way is not so much.

36:34That's why it's surprising, I guess. I guess I've been so steeped in this new way, I'd forgotten the old way that nothing happens until PR.

36:41Amelia Wattenberger:If you're talking about agents or other... Like, do you work with other developers before? No, I mean, a lot of it is not collaborative. I mean, in open source, yes, other developers, but that's usually the potato over the wall. It's not a lot of like team collaboration getting to X. It's more like, you know, solo getting there. Which is weird, right? Like it is like, especially if we work on, I know as a lone wolf developer, I kind of like it. But like things should be shareable, right? Right. Like, especially teams working on the same feature. We're doing all this like duplicate conversations in Slack that like explaining the code that we're writing locally.

37:24Amelia Wattenberger:And like there have been a few stabs at it and Replit kind of does this a little bit. But like by and large, you know, typically a project is like I have a Slack conversation. I have a linear issue. I have a GitHub issue. Maybe there's a century bug. I look at it I talk about it maybe I put comments on a doc in Notion and then I make a branch in my like local VS code and then I write code on it and you know the outputs in the browser and then like oh wait I have to work on this other thing so I like commit it and I put the branch down and then I like go work on you know feature b there's a like you know there's a bug in prod oh no let's fix it and then I like go back to my future branch but it's like wait what was I working on and like what's the progress and what were the conversations I was having around this um and like the proposal is just like let's have a new primitive which is a workspace where everything lives so that it's easy for me to put down and pick up and hand off and share And like our work as developers is only partially code.

38:35Amelia Wattenberger:A lot of it is just like context integration and testing and all these other things. And so like, let's build a primitive that works with that and stop pretending that everything's code all the time. All right. So that project was at GitHub Next. And now it feels like I'm just watching things and like collaboration is even more important now because the same tools that would make things collaborative with other teammates feel like the same tools that we need to work with agents where we kind of just need like a room for everything for a specific task so that like the agents all know what each other are working on and I know who made what change and like we're having conversations in in public where like I'm talking to agent a and agent b knows about it and we're all working off like the same spec and documents and, you know, like embedded web browser.

39:30Amelia Wattenberger:So that, that was kind of the like core, the spiritual core of this new app that I've been working with augment code on called intent where let's, let's like add this primitive of a workspace. And then now that we have that, so within the app, uh, everything's a workspace. It's not like an agent chat. Um, anytime you want to start a new task you would create a new workspace so you say for this repo i'm gonna add i keep my demo i keep doing is adding dark mode to my website right um and then you know that has its own like get work tree it has its own set of agents it has its own um set of like rich uh markdown documents to explain all the context and the spec and then because that's all isolated i can like switch between adding dark mode to my site and like maybe adding a new blog post.

40:26Amelia Wattenberger:Right. And those can have agents like whole teams, swarms, whatever orchestrations of agents going at the same time. And they don't get confused between each other. And then when I switch between the workspaces, I have the visual reminder of like maybe the browser is open or the spec is open and you know like exactly like what the progress is and who's working on what. so it's a little bit of like a redesign of how do we want to work and can we uh build an environment where that's easier and then to come back around to what we were talking about it just makes it so much easier to work on you know the 10 things i'm working on at once like this last weekend i was adding you know 10 separate features to this app and it it's still a little bit crazy making but it's so much easier.

41:14Well, the challenge, I think what you may have solved for, and if I, if I understand correctly, a workspace probably has its own Git work tree, which means it's not the same shared code base. It's a, not really a true GitHub fork kind of thing or a Git fork thing, but it's literally, it's, it's a separate instantiation of the code base with its own Git history that it pushes probably to, to main or to a branch. If you're branching, And that's exactly, you know, I was just thinking about this this week, really, with AgentFlow. It was like the next way to get to where you want to go, I don't want to swarm a bunch of agents, but I do want to work in parallel.

41:51I want to have a couple at least because there's things that might require a CLI touch and an API touch, which are two different code bases. I want to keep the CLI maybe open source and the API is proprietary. Maybe it's source available. so you've got different layers to your product. While one agent's working on the CLI, the other one's working on the API, I'm actually working in a model repo that has multiple sub-repos because I'm trying to keep the agent in one repo kind of thing. Which I think now, Codex actually happily just goes wherever it wants to. Whereas Claude seemed to be like, you know what, I'm only here, okay?

42:29If you tell me to go over there, I'm going to get confused. This is me right here, wherever you instantiated me at. and that's a new word, instantiated me. Instantiated. I like it. You know, just this idea, though, that that's spot on. That's exactly where it needs to be because if I'm going to have more than one agent, we have to have more than one code base. If they're stomping each other's toes, it's like two devs working on the same repo trying to commit the same. It's like, no, that is not how it should be. How did you, I mean, it seems obvious, but how did you get to that realization from a technical standpoint also from a design standpoint

43:09Amelia Wattenberger:mostly just trying to constantly pushing the bounds but like they call it the bleeding edge because it's you know painful right like trying to use all the new tools and like i you know i found myself running eight different agents at the same time all working on different tasks and like they step on each other's toes and like they start accusing you of changing things right they're like yeah you gotta stop changing this file there's changes in that file i didn't do just so you know okay like it wasn't me he's very apologetic i'm not committing that i can't own that i didn't do that i got that one yesterday by the way that was pretty funny really yeah i was like okay chill i'm with you i know the other agent did it i should be smarter and have you know intent here and do you know multiple workspaces and be smarter about them so i'm just not there yet okay stop yelling i'm coming yeah it is interesting like the so two things two things about what what you're talking about um when it's like a lot of the like new explorations from the other tools are around a work tree per agent thread so each agent has has its own Git work tree.

44:33Amelia Wattenberger:And that's actually not the direction we went down. We have one work tree per task. And there's kind of these pros and cons for multiple agents working on the same tasks within the same environment. That's kind of been like an interesting research exploration. And I find it actually works quite well where as long as the agents know what each other is doing and are able to see like, okay, These are the three other agents within the workspace. And like this one's working on subtask A and I'm working on subtask B. You really don't run into issues. The one set of issues that we did see a lot was like we have them auto committing.

45:15Amelia Wattenberger:It's like a setting. So like you split a task into subtasks and then you like you say like you three tackle these three subtasks like in parallel and then auto commit when you're done. And because Git is like the way Git works is like you have to stage files or hunks and then write a commit message and like run the commands like that. They would step on each other's toes with that. So we added a tool for like queuing auto commits. And other than that, like you get all of the benefits of like, you know, one's working on creating a component. The other one's working on adding the component to the layout or another one's working on like global style changes.

45:55Amelia Wattenberger:Right. Like you do want some of these tasks to be done within the same environment, because like usually when you split things into tasks, they're inherently linked together. And like, you know, API change as well as like another one's making the page. Right. Like you need those to be in the same place. And like you want visibility between those agents. You don't want just like completely new cloud environment for all of these agents as they're working on the same thing. So like it's been interesting exploring, like, what do you want sandboxed? at any given time. And then the other thing that's been really interesting is just like orchestration paradigms, right?

46:32Like, yes, with our given tools, it's hard.

46:36Amelia Wattenberger:It's hard to explore any kind of orchestration. Like I actually had like a home cooked, like parallel running system. Cause I was doing large refactors where I was like, I'm going to make a markdown file and it's going to like be split into 10 tasks. And then I'm going to kick off like 10 agents in parallel because it works faster and I can use cheaper agents when it's all spelled out. And like, it just like was not conducive to like actually exploring like how to make that really nice. But like if you have the primitives to like kick off different types of agents and they all have visibility into what each other is doing, there's some really interesting paradigms.

47:18Amelia Wattenberger:So like our default paradigm within the app is a workspace starts with a coordinator agent, which writes the spec, and then it delegates subtasks to implementers in waves, and then it asks a verifier to validate those changes. And it works really well. But there's other paradigms like spin up two agents and they ping pong between each other. And then agent A goes and it critiques what agent B did and does its own thing. And they level each other up as they go. And there's all these really interesting paradigms that i like i really want to explore and then like you look at gas town right and there's like you know like there's a deacon and there's you know it's a whole town yeah yeah i think there's mayors and stuff like that citizens i don't even know it's like a boot i didn't go that deep yet i um but is it like that's exactly it right it's like impossible to follow when you're just like starting from the the primitives which are like an agent um so So there's like a lot of really interesting things as far as orchestration that I feel like are just now being explored.

48:31Amelia Wattenberger:And there's a lot more that can be done there. But I think we just need the tooling to make it easy to understand who's doing what and et cetera. Take me into this exploration around coordination. Because I feel like that's probably the hardest part once you have multiple agents. I do think it's smart to do a task-based versus agent-based because you may want to have multiple agents. But then you have the issue of stomping on toes. So you have to kind of coordinate it. Do you have a coordinator or an orchestrator? Or do you kind of give them hats? Do you give agents like, hey, agent, you're the verifier now.

49:14And now they assume this role. Is it stuff like that? How does the agent know? And then how do you trust the model to let it explore and push back on the other agents, I guess, thoughts, if that's how you want to frame it?

49:30Amelia Wattenberger:Yeah, yeah, that's a good question. So we have the concept of a specialist, which is just it's like a persona, right? And it's like a prompt and a default model and set of tools. so we ship the app with a default set of specialists like we have a coordinator an implementer a verifier there's like a pr reviewer ui designer there's some just like things to get people started but users can tweak those or make their own specialists um and the default flow we have that we we've found actually works really well is one coordinator that you start a space with and it does all the research and it like breaks the spec and it makes sure like it's usually the main thing that you're working with as a developer and then that coordinator on the fly can create other agents like either blank ones or specialists so we in our default prompt for the coordinator we say for the subtasks delegate them to implementer agents and implementer agents um depending on how token sensitive you are like it can be a cheaper model or it can be you know more expensive i really want this to be as robust as possible and like those implementers get their own sub task um and it's the note that they can like write in and read from And like, so they like, it's like kind of scoped.

51:06It's really well scoped for them, but they're also told who else is working within the workspace.

51:12Amelia Wattenberger:And I feel like that's made a lot of difference is just like the agents know and have visibility into who else is working. Right. Like it would be weird to join a team and like not know who else was on the team or what they were working on. Right. Like I'd be like, I don't know what's the scope of my work. so it's just like this really smooth flow of like coordinator delegates to implementers and then we also found that agents don't always do exactly what they say they did and like maybe this will change but for now sometimes an agent would be like i did xyz and then you know you'd look at it and they only did xy and so the coordinator is also told to delegate a verifier agent, which again, could be a cheaper model.

52:01It could not be, um, you can,

52:03Amelia Wattenberger:you could use like a codex model if you feel like that works better for this, this case. Um, and that, that verifier just knows exactly what the work is that just completed. And it, it like goes through and verifies it, you know, it runs the tests, it, uh, runs the app and looks at it and, uh, keeps everyone like to task and it reports bets of the coordinator. And it's actually kind of a really interesting flow. I keep meaning to post about it of like having this one chief of staff that you talk to for everything. And like it delegates to other agents. So it's like an interesting, um, it's a different flow where like the work can be happening and I can talk to the main agent about the work that's happening or like, oh, I found a bug and I'll just like, as I go through the output, I'll keep telling the bugs to the coordinator.

52:55Amelia Wattenberger:And because it's not doing the work. Sending messages is not like queuing messages because they get sent right away. And it's not like interrupting an agent because the work doesn't stop. So it's been kind of an interesting change in pace of how I do the work with agents. Yeah, that's exactly where I've landed too. I mean, it's funny how we think like this because I basically said to the agent, your work here is too important for you to do the work. You're the planner. You're my orchestrator. you check on things, you confirm the task is clear. We sub-agent that. And because, unfortunately, Codex isn't smart enough to have sub-agents yet, or at least that I'm not aware of, and I just haven't explored that, let's open a new tab.

53:41And we create a flow, essentially, and I'm calling them flows, where I want to get to, you know, from 0.5 to 1 or whatever phrase you want to apply to that. I want to have this task done. But you're my planner. You're the orchestrator even between a couple of repos. So in the case of the sandbox, I have the main repo where I've got, I thought we actually landed on the fact that I just need like one primary operator because I was trying to have an operator for that repo and an operator over here. And I'm like, well, hang on a second. Now they don't know what you're doing. You have all the context and it got messy.

54:16and so even with one operator between two repos and sort of two different versions like you're here to battle test this thing and they're over there coordinating work on the product to evolve it and so it just made a lot more sense they're like it got confusing for them and for me even so i'm like your your work here is too important to do the actual work you're my planner you're my verifier. We are real time. If we see a bug, we log a bug kind of thing. If we have a thought about something, we'll spike out an idea or drop a document in there. But really it came down to that as like, I want to have sub agents, they go and do the work.

54:56And yeah, you're right. If you had sub agents or multiple agents working in a room together and they weren't told to introduce themselves, that would be weird. That would be weird, right? Like, well, hang on a second. and there's somebody touching my files over here. There should be nobody in this room. And, you know, they've got no eyes and no hands, so to speak. So we kind of have to give our agents to do this work. We kind of have to give them eyes to see something and give them hands to maneuver something. And at the same time, I've learned, tell me if you've heard this before, if you feel this way at least when I say this, trust the model.

55:30Does that ring a bell or resonate anything inside, Jamilia? Trust the model.

55:36Amelia Wattenberger:It makes me uncomfortable. It makes you uncomfortable? I have the opposite philosophy. Okay, gosh, let's dig into that then. So I've pushed on the model to not trust it. And I don't know about you, but lately I've been just specifying things I want to do so well that when they come back with feelings, I guess, I don't know how to, their intentions or their thoughts on what I'm putting out there. A lot of my responses end up being, I like that. What are your suggestions for this? I review their suggestions. And I'm like, my, a lot of my responses is some idea. They go and turn on it. I review it and I'm like, this is awesome.

56:15And then I ask it for clarity and blind spots. And like, they go back and review what we've done. Cause I don't got time to do that. Go and review it for clarity. Is this clear? Are there any blind spots? We're not thinking about that. If we started to implement this now, the agent could do it, but it would be guessing along the way, you know, and then that's not my intent. That's not my intention. And so then they come back like, well, here's five or six things that is unclear. And here's like six major, here's two high blind spots and here's two sort of mediocre ones. And so I can think about those for sure, but I'm busy podcasting, responding.

56:54Like I have a job. My only job isn't this. If this was my only job, I would probably read a lot more of it. So then my next question is to the agent, what are your suggestions on those things? because I want to review your suggestions because you've got the time to examine it, agent, not me. And so that comes back with these suggestions. I read those. I'm like, yeah, those are great suggestions. My next response, Amelia, is what? Do it. So a lot of my interaction with the model is in a way, trust the model. Not because I blindly trust it, but because I've learned that it's got some pretty good thoughts and it's a heck of a lot smarter than me and certainly way faster.

57:34And so a lot of it is my intention, my ideas, my desires, my problems, it's helped me solve. And my responses are pretty mediocre. It's like, what do you think about that? And do it. If you look at my history, my up history is like a lot of that. Like, what do you think about that? Do it. What do you think about that? Do it. You know, I don't know. Anyways, maybe I'm revealing too many of my cards here in terms of like just how much I do trust the model when it comes to what I'm making.

58:03Amelia Wattenberger:No, that's exactly right. There's, um, augment launched this feature, uh, a while ago. Um, and a lot of these agents have this, which is enhanced prompt, um, which is like you write a little bit and then you ask it to enhance the prompt and it turns your, you know, 10 words into a hundred words. and like I like this because this is in a lot of ways what we're doing right we're we're trying to have a compressed version and this is what talking is right like you have a maximally compressed version of what it is that you're trying to represent that lives in your head or maybe doesn't even fully live in your head it's just like add dark mode to my site right like and this wouldn't work previously with code because there's a lot of things that are inherent and you would have to assume in order to turn those words into an implementation right like first of all what is dark mode right like what does this person mean when they say dark mode or like what pages should this apply to should this have like some kind of affordance to switch between dark mode and light mode.

59:19Um, and there's like a set of like less and less obvious

59:25Amelia Wattenberger:assumptions that you can make about what the person means when they say these words and like you're doing, you know, I'm not going to bring out numbers again, but you're like, you're extrapolating. Please do. I like your numbers. Don't feel bad about it. You're going from 0.75 to one. Okay. From add dark mode to, this is the code for adding dark mode. I feel like enhance prompt goes from zero to 0.1. Where like, it's a little, it's the negative TLDR, right? Like it's a little bit more like, this is what you mean when you say that. And then the way enhance prompt works, it's really nice because you can edit it, right?

1:00:04Amelia Wattenberger:You can read through, okay, this is a little bit expanded and like, oh, you're going to assume I mean this. Let me actually tweak that because I see it and like, it's actually not what I want. and then like extrapolate a little bit more you get to this like a spec right like that's like the spectrum development um where like a spec is supposed to be somewhere further along that spectrum of like let's add more detail into what that would mean in the code but we're not going to have the fully fledged implementation because that would be the code right um or like that would be hard to read through and you know I can go from 0.5 to 1 and make assumptions like those assumptions along the way let's like get out the like big decisions and like agree on them before we do the work so like that's kind of how I think about the way we work with these language models is like we say a little bit and then they they basically flesh it out for us and turn it into code.

1:01:13This episode is brought to you by our friends at NordLair. Here's a weird asymmetry in the security world. The tools that protect the fortune 500s, the big guys, the ones that actually stop ransomware before it encrypts everything. They aren't available to most companies that need them most. You, me, small businesses, midsize teams, startups. They get breached constantly we see this every single day in our news cycles and the reason usually isn't that they're careless it's that the enterprise grade stack assumed that you had a dedicated security team to run it most small businesses medium businesses startups they just don't have this nord layer is closing that gap it's a network security platform that combines vpn access control and threat protection into one place built on zero trust deployable in under 10 minutes no hardware no quarter-long rollout and friends this is affordable plans for just eight bucks a month enterprise-grade security minus the enterprise size team needed to run it get up to 22 off norway your yearly plans plus an extra 10 off with the coupon code changelog hyphen 10 hyphen NordLayer.

1:02:33Yes, the code is changelog-10-NordLayer or hyphen, whichever you choose. Try it risk-free with a 14-day money-back guarantee. Check it out at nordlayer.com slash the changelog. And we use our link. As you know, you're supporting our show. We appreciate that. Once again, nordlayer.com slash the changelog.

1:03:01Thank you. My audience knows this. Love Augment. I use Augie on the daily. Augie is my one-shot tool. I love to just give Augie something just to churn on for a while. Like I'll do a lot of thinking with Cloud, with Codex. I'm no fan of just one particular. I feel like any developer really should use everything they can actually afford and get their hands on. um and i'll even just let me mention you mentioned uh what how did you describe you said token i don't think you said token efficiency but uh where you want it to be what'd you say

1:03:38Amelia Wattenberger:do you remember what you said i mentioned before like people are differently token sensitive token sensitive i like that phrase a lot i feel like that needs to be you know a phrase in our world because there are some that are more token sensitive i'm like hey listen when we're in when we're orchestrated while you're telling this agent what to do we're going to use codex or a gpt 5.3 codex like on high we don't care like i want the best code we can i have plenty of usage space so don't you know take me down to a lesser model and i don't care about speed i just you know i'm it's working in the background so i'm not trying to get there in two minutes when five minutes is better or whatever the case might be.

1:04:19So being token sensitive, I think, is really an interesting place to be. And I'm just sort of like, you know what? I have access. I'm going to use access. And maybe somebody is a little less like that. I'm trying to figure out where I was trying to go with that because I kind of sidetracked myself getting on token sensitivity there.

1:04:40Amelia Wattenberger:You're talking about like enhanced prompt. Yes. How much I like Augie. Yeah, my one shot. Yeah. Yeah. Thank you for bringing me back there. I really appreciate that. So big fan of Augie. Now, this is the feature I've never used, though. I've never used this prompt enhance and so I'm going to have to try it. And I guess my challenge there is that when we put our little pros into the text box, the CLI, the web app, I think Codex now has an app version, which felt so non-hacker, so I had to go back to terminal. Even though they gave me 2x usage, I'm like, no, no, no. I'm going back to where I feel hacker, okay?

1:05:25CLI is just where it's supposed to be. Don't take us away from that. I guess maybe intent did that too. So sorry about that.

1:05:31Amelia Wattenberger:It's okay. Maybe you have some thoughts on that front there. I feel like hacker is the way don't take me too, too far away from hacker. I feel at home in the terminal, you know, it's about feeling, feeling good. But here I am in the terminal and I just, you know, I have no affinity to like which model I'm using or which platform I'm using, but I've never used this, this enhancement. Cause I feel like at least for me and why I haven't leveraged it is because part of it is this dance. Part of it is this trust the model. And I'm not just trying to get to software. I'm trying to get to reasoning for myself, not just for it.

1:06:07Even though I say, hey, what suggestions do you have? Do it. That whole mantra back a minute or two ago, despite saying that, I still want to learn along the way. And so part of that learning is my troglodyte prose into it that comes out the other end with clarity. And it's that back and forth that I feel like it's like a potter and clay in a way it's like the it doesn't turn into the bowl right away or this beautiful thing like for example here this my coffee cup is it's a hansami it's japanese like this is japanese clay right here amelia it's just a typical coffee cup but somebody made this with their hand and it's amazing i love that to get to that you have to feel all the bumps you have to feel all the pressure.

1:06:55And so I haven't wanted to enhance my prompts because I feel like, not that I'm against it, but me sort of feeling the pace, feeling the pressure, the back pressure of my dumb idea turned into a good idea to understanding that it's a good idea, to understand a new paradigm, to a different way of thinking about it, maybe introducing bun versus go or svelte with svelte kit versus something else versus a tan stack start. Like, oh my gosh, the war's out there, right? And I like, the wars are immense. But I feel like that's where I'm personally learning. And maybe that's a today problem, which I think is really important to mention, right?

1:07:32This is 2026 and this is how I feel today. This is the problem of today. Maybe in six months from now, I won't care so much because the agent is just so smart. I should just forget trying to understand. I don't know. What do you think?

1:07:46Amelia Wattenberger:I have a lot of thoughts there and I'm trying to think of like, what would actually be most interesting? like I feel like as developers we're kind of having an identity crisis right where like are we potters and are we supposed to be opening like a pottery factory and like that's you know like there's two main joys that I feel with being a developer one is like the craft of the code and learning the tools and knowing what are the bricks that you can build your buildings with and the other is just like building things and like thinking of something and then it exists right and I think the way you might feel about this new wave of tools lies along those lines of like which is the greater joy for you and what do you find the most meaning from sure and I've always been like which is why I've moved more and more towards design, you know, like I started as a developer and now recently I've had more and more design roles, um, is that it's just so interesting to be able to imagine something and, and then it like comes to life.

1:09:02Yeah.

1:09:02Amelia Wattenberger:Um, like that is amazing to me. Um, and like a lot of these are things that did not exist before, you know, like whether they're good or bad, this is the first time that they exist and there's something like really amazing in that and um a lot of i feel like your your main motivations and what you find the meaning in it informs the way you use these tools um that's true and i i do think it's important to still feel the clay right like you still want to be in touch with the physics of what it is that you're working on and you want to feel in charge and in control and not like like I think about um art right and like how generative art is changing and like people are kind of having to think about like what is art and like what do we find meaningful and like um if I type a prompt into a box and something comes out and it's like the Mona Lisa, right?

1:10:03Amelia Wattenberger:Which honestly I think is overrated, but it's a good example. It like, how does that change how I value the, the, the thing that came out of the box? And like, can we have workflows where, um, the thing that we think is meaningful, which is usually like effort and thought and like the humanness that goes into it. Um, can we, uh, leverage Gen AI to make tools for which people to like make a thing and then like really sculpt it and craft it and like put intention and effort and human feelings behind it. So that it's not just like type prompt into box and it spits something out. But it's not the usual like paint that we work with as painters, right?

1:10:55Amelia Wattenberger:And I feel like this applies to code too, where like, how do we leverage AI to act like a tool and not like just an automation of what it is that we do as developers? And I think that's like a little bit for me what you're getting at. Very much, yeah. Like, what do we want out of this really? Yeah, it's like trying to build a house and you're framing and instead of having one frame or with one nail gun you have infinite in a sense or unlimited nails or just something that not so much speeds it up it's it speeds up the idea of thought you know you'd mentioned um as a designer i feel like i'm a software designer now i'm not a software developer in a way yeah uh i'm designing it rather than developing it but it's kind of a both but it's more design than development.

1:11:49And I think you're spot on with the identity crisis. I think a lot of people even listen to this, probably somebody not listening now because they got upset midway, you know, maybe you're mentioned of more numbers potentially. I'm just kidding. I love that. Maybe it's something I said, you know, but I think we're all having this identity crisis to what are we really trying to do here? And I, I think you're right too. It depends on your motivation and how you leverage the tool. I think the way I look at what AI is today for developers and how I'm using it is very much like a tool, not a create this thing for me.

1:12:29It's helped me create this thing. Help me use your infinite knowledge so I can connect my own dots, you know my own problems is why i keep inventing new software um my wife is i think she might be on the edge of being not cool with it anymore we'll see because i will often say to her in the morning or in the evening we go to bed babe i invented something new today and this is a very common occurrence like i mentioned gauntlet and then there's agent flow and then there's uh bkt A lot of these things are acronyms for some reason. My son's like, Dad, why are you naming all these things acronyms? And it's not really intentional, I guess.

1:13:14They're not really acronyms. But AgentFlow is an acronym. It's a full phrase. So get off my back. But although the command for AgentFlow is AF. Amelia, how cool is AF, right? As a command, you type into a prompt.

1:13:28Amelia Wattenberger:I think you win at naming things. Yeah. Well, thank you. It's not the millennial version of AF. It's just kind of cool either way. It's super short, super slim. Hey, millennials start with school. I'm an older fella, so I think I'm Gen Z, Gen X. I think I'm Gen X, if I recall correctly. I forget which Gens there are. I just know that I'm old now. I feel like I still feel young, but I am old. So I'm often like in a room with folks. I'm like, yeah, you're kind of like 10 years older than we are. I'm like, well, sorry about that. Just maybe 16 years older than you are maybe even. You underaged me.

1:14:02Thank you very much. it's weird how that happens but very much in this identity crisis very much trying to use ai as a tool to help me create and invent and design new things versus simply magic box give me something uh you know everything i know nothing thank you very much software created product made open source done you know 10 million arr you know like that's not my that's not my world you know I'm not even trying to vibe my way to get there. And I empathize with those trying to get there because the world, the world says be richer. The world says make more. The world says ARR, right? The world says high valuation.

1:14:43The world says raise$100 million. And kudos to those folks there. By the way, Michael Greenish, congrats on the 100 mil for WorkOS. Big fan of Michael and WorkOS, but they just raised their 100 mil just yesterday. And he's coming on the podcast this Thursday. So it'll come to the actual airwaves here soon. But no knock against that. That's cool. And he started WorkOS before the advent of AI. And now AI is sort of reinventing or helping folks reinvent their platforms. Because, well, now, Amelia, you don't have to just support Go as your SDK because you're a Go shop. Now you can support TypeScript.

1:15:20God bless you. Go. Pick your language that you want to. maybe even Ruby again, who knows, because that's the fringe. That's the edge case. What are your thoughts on that? It's like not just a tool to make more, but a tool to sort of flatten things because rather than just support an interface for my software, an API for my software that is only an API in this way, now I can have an MCP server, I can have SDKs, and I can actually have good SDKs because we can move a little faster and support more of a footprint. that the footprint can widen because we have this new invention tool, this new design tool we have?

1:16:03Amelia Wattenberger:I feel like my answer isn't super different than what we've talked about before where it's like now that we can build anything, what is it that we built, right? Like it's so tempting to add 10 new features to an app, especially with you get the 70 % really quickly. like why don't you just polish that up and ship it out and you know you're done you know your your app does you know it's an ide and it's your notion and it's linear um but then you just have this massive app and like why why yeah right like you can only have so much intention behind what you build that it's on it it feels like it's becoming um the question is what you don't build as opposed to what you do build right like if we if we can virtually eliminate our backlog in a week how does that change what we decide to ship because the answer can't be ship everything yeah because then you know your stuff or it doesn't stand for anything right if it stands for everything so it's been as someone who likes to do scope creep because i like like fully fledged versions of things and like really thinking through you know what is what is it if it does everything um it's become like it it's a new muscle that i have to exercise right restraining know to things yeah for sure your choice is your superpower amelia do you uh are you a fan of notion because i think there's some change in the notion world which is interesting i just saw something you thought notion was just docs everything just changed and so now they're like a version of clod i guess in a way where they have notion agents and they're a sponsor i'm not sure they're sponsoring this podcast and they may get upset but that's just how it works but I think I'm kind of confused by Notion because like I like Notion.

1:18:09I use Notion for what it has been, let's just say, not what it was, what it has been, which is a place to organize teams of folks or one people or many and collaborate with external folks on documents, workflows, statuses. It's a version of a document state machine, right? Comments can happen there. You can invite people in there. and I think what inhibited my usage of it is just the difficulty it is to manage that kind of workspace anyways so kudos to them for throwing agents in there uh or notion ai which I use and I think it's super awesome but now I'm not sure what notion is because if notion is where I go to build agents that don't even belong in notion what is notion so it kind of comes this comes this notion or to this thought that you had here, which is, you know, if we have, if we have no restraints and we could just put, we could just build anything, what should we build?

1:19:07I'm paraphrasing what you said, and I don't remember perfectly my brain and feel free to restate your, your phrase, but how do you feel about that? I suppose, where you've got a tool that was sort of known for one thing and now it's almost reinventing itself to the point where you're like, wait now I go to Notion for the thing I didn't go to Notion for and I'm just using them as an example because it's the most clear recent example of a not so much a major reinvention but a major major new feature that sort of makes you rethink and they even said in their advertisements on who they are we're not this anymore you thought we were just this now we're 60 percent not just the 40 percent you thought we were before so what do you think about that with this massive of reinvention and just doing more than you should could.

1:19:56Yeah.

1:19:56Amelia Wattenberger:I think it's like this reckoning of how do we use computers? There's, there's two analogies that I have in my head. One is like the IDE was invented almost half a century ago. Right. And it's very good at working with the primitive of code and you're always working at the code level, right? You're reading the code, you're writing the code, you're clicking between files. And we like through AI is chat on like on the side. And we're like, okay, these agents are going to write and read the code for you.

1:20:35Amelia Wattenberger:So that's like, there's like step one, which is like, what are the primitives build the UI around the primitives? Step two is throw AI so it can also use those primitives. And then step three feels like kind of like reinvent the tool now that um our role in the system is changing right where we're no longer mainly working with the primitives we're like moving up a level on the ladder of abstraction and focusing more on like uh for code the meta of like what's the intent how like what are we building what are we changing about it what exists in the code base um and it It feels like Notion's kind of going through the same, okay, we had documents and then we threw AI on the side of those documents.

1:21:23Amelia Wattenberger:And now they're kind of moving to step three, which is like, let's work on the meta, right? Like I have this long list of bugs in my app. Can you transform that into like a summary or into some like higher level representation of this so that I can interact with it? in like a much higher leverage way. And then there's like a next step, which I'm super excited about adding to intent, which is you can go up another level, which is like, if you add AI at a higher level, which I guess is kind of what Notion is doing, where like, you can talk to an agent about all your workspaces. And you can say like, hey, what are my last 10 PRs?

1:22:12Amelia Wattenberger:Can you make a workspace for each one of these. And then you're like, what happened in my workspaces last night? And oh, like move, like nudge along these three tasks, right? Like we've gone another level on the up on the ladder of abstraction where like, we're, it's like, um, moving higher in the air where the things on the ground, there's more of them, but they're smaller. And like, they're individually like taking up less of like our view as far as like our perspective is moving up and up and we can do more and more but we can't use the same representations that we have then and I feel like a lot of software is just going along this this trend of like okay like the way we interact with this is changing and how do we design around that and how does that change our identity as software yeah i like that idea i mean that's exactly it's exactly what i want you know as i mentioned before i confided in you and to the to the rest of the audience that i have a hoarding problem with my software and i can't stop inventing new things because i keep finding more problems that i want some else to solve that i'm like well why don't you just take a stab at it because what's the worst thing can happen like will you fail to yourself like what's the big deal let's just do it right right and it's the tension between like i want to feel the medium still right and i want to do more things right so we need interfaces that make it possible to move between those two levels i think about maps a lot right like a map is like an amazing feat of design like google maps right we have static maps but like google maps right like it's like first of all it's just amazing that we've mapped you know basically everything in the world but like if you think about all of the design decisions that go into like think of the map of your town at a high level you see all the highways and like what's green and what's not green where are the busy areas you zoom zoom in a little bit you start seeing like smaller streets you zoom in more and you see like paths and you see buildings and you zoom in more you can even go inside a building like Like, it's just like amazing the like perspective shift that they have to manage in like one interface.

1:24:26Amelia Wattenberger:And like the fact that you can seamlessly zoom in from like inside a building to like a whole town to like a whole country. Like it blows my mind. And I keep thinking about that. It's like, how do we design other things that way now that qualitative things are a little bit more malleable? Yeah. Yeah, I kind of want to either have, own, or create a software factory, not so I can churn out software, but because I think, and maybe at some point everyone else will truly be a builder because this will just flatten. But I think for now, it's the was developers to be software designers or whatever our identity crisis is leading us towards.

1:25:09I really want to have a software factory, but like I said, not because I just want to churn out software. I want to have the zoom in. I want to have the macro and the micro because I want to have the big picture designer idea. I want to have the intent idea. I think the name intent by Augment was just spot on. I don't know if you were involved in the naming of that, but was that your name? Did you name it?

1:25:31Amelia Wattenberger:That was not me. Okay. Were you in the work for the scrum or whatever you want to call it to sort of think through the name? Yeah, we kicked out around a lot of names. It started out as workspaces, which is very literal. A little, yeah, a little bland. You hear it and you're like, okay, I kind of get what that is. But like we kicked around a bunch and intent. Yeah. I still like it, which means a lot. I think intent is a pretty good name for where it's at. I'm not sure because even with my own names, I'm like, I don't know. You know, I don't know. All of the names you've mentioned have been grade A.

1:26:11Well, thank you. I think it's fun as a designer, as someone with intent and good taste. I think I have good taste. I've learned to trust my taste because I keep solving my own problems and I think there's this agent psychosis thing that I think people have talked about and I'm like Adam do you have agent psychosis because all your ideas are good but it's not the agent telling me my ideas are good it's me making the thing and using the thing and be like okay the idea is good and so I don't feel like I have this agent psychosis although I am very aware that it is a phenomenon happening to folks, but I do want to zoom in because I want the understanding and reasoning.

1:26:53I do want to understand my choice between, so here's something I deliberated between recently, which is, I like languages that compile to a binary. So Rust, Go, Button. And then there might be more, but I'm just not familiar with every language it does. And so I largely deliver most of my software in Go for those reasons. It's just easier to deliver a binary, right? So easy. Rust is a harder language to reason about. So I like Rust. I like to use Rust when it makes sense, but largely my language of choice for most things is Go. And then I recently was building something with Ruby because I really want to get back into Ruby and particularly like Ruby on Rails, because Rails does APIs very well.

1:27:38And it obviously does web UIs extremely well, right? It's got models, views, context. It has all that stuff. Sorry, controller, not context. but then I was like, you know, I got to a certain point with this application where Ruby on Rails could have done it, but then it posed some challenges to deliver us. I won't go into the details, but you can't compile a Ruby on Rails app to a binary. It's just not possible. And I'm not going to solve that problem yet. I don't think I'm ever going to solve that problem. Maybe somebody else is trying to. So then you have Bun and you're like, well, Bun can deliver me a TypeScript application compiled down.

1:28:16and because the way SvelteKit works and Svelte works, it's just writing, it's an abstraction above plain old JavaScript, which I don't want to write because who wants to do that anymore, right? It's not smart. It's like not using agents. So I want to use SvelteKit and I want to make some really good front-end UI and just different things in the JavaScript ecosystem and I want to leverage NPM and all the goodies that are there but I can still have a compiled binary, right? I can take a Svelkin app inside a bun and compile down to one thing. And so all that to say is like, you know, when I explore my world, I want that.

1:28:53I want that micro. But then I want to zoom out to the macro and say, show me the map. Where's the United States? Where's Earth? And Earth is like all my projects. And so I very much as a software designer, developer, builder, identity crisis, whatever you want to call me or us. I want the micro and I want the macro so I applaud you for that work on intent to just give the zoom in and the zoom out it's very much important yeah this is actually like it's something

1:29:23Amelia Wattenberger:I've grappled with for a long time is like how do you represent something at the macro how do you represent a code base at the macro and it's a really really hard problem for many reasons uh one is which like there's no one representation that will satisfy everyone because we all have different mental maps of a code base um and it there is no like you either visualize it as like this is the directory structure because that's the only like pseudo spatial thing that we have or like you use the ast to generate a map um or you do something more semantic that's like in my head the structure is like here's the front end here's the back end here are the API routes um so there's just like a lot of there's a huge exploration area as far as like what is a way to visualize code bases and tasks and like monorepos and sets of code bases and like how do you zoom out and what does that look like and what are the affordances that I one at that level.

1:30:37Amelia Wattenberger:So it's something that I'm very excited to explore more, but has also been virtually unexplored. So there's a lot we can do in the short term and in the long term. I wonder, Amelia, if we're just code explorers in the end, you know, if we just become that, just become code explorers. I don't know. Every time I hit a hurdle, I want to explore further. And then I got to leverage my taste and justification for should I build that or should I pull back? Should I do that? Should I add this feature? Should I not add this feature? So I very much feel the tension that you described here. What is the best way to get started with intent?

1:31:19Is it just if you had to give someone an on-ramp, go here, do this. What would it be? You built the thing. Show off how you're proud.

1:31:27Amelia Wattenberger:Come use it. We love feedback. it's free bring your own agent you can use cloud code or codex or we recommend augments agent because it's really good at staying grounded especially in complex code bases augmentcode.com slash intent and download it we have a mac release right now and we've got windows and linux in the works and just like open it up start tell it what task you want to do and it'll make a spec for you and go from there. It's been really fun playing with it. And it's like, it should be a very, very smooth, fast onboarding where you just hook it up to your assisting CLI agent. Yeah, I love that.

1:32:12I already had Augie installed it myself. And so I was like, yep, sweet. Already got that. I chose a repo to play in. I started playing with it. Of course, I went straight to my micro version of it, which was just me and an agent versus the swarm. I know we can go right or left, which I think is kind of cool. You can bifurcate the direction you want to go to there. But I was like, I'm comfortable here for now. Although I do think I want to swarm or have multiple agents. I forget what the terminology was that you used, but it was something like swarm or multi-agent or something like that.

1:32:43Amelia Wattenberger:So for any given workspace, you can choose the specialist that you start with. And so you either choose the coordinator, which does all this orchestration, or you can just do like one agent that doesn't delegate. It just writes a spec for you. And then you ask it to get more of a back and forth. Yeah, that's where I'm at for now. I'm still getting comfortable with trusting the models and trusting the agents to do all my things. But I do like the exploration of this world. Anything in closing, Emily, anything I didn't ask you that we can tell out. How can people track you, trace you, follow you, pay attention to what you're putting down?

1:33:22I know you have a blog. Do you still write a lot? Are you, you know, how do you feel about writing? How should people keep up with you? What's the best way?

1:33:30Amelia Wattenberger:I have so many thoughts and yet so little time to write them down. I do have a blog. If you search, if you can spell my last name, you can find me on all the things. W-A-T-T-E-N-B-E-R-G-E-R.com. Nailed it. Exactly. I had it written right here. I was on your website. Yep. Wattenberger. I have monopolized my family name online. Sorry, everybody in my family. Well, you know, when you're famous, you're famous, right, Amelia? Yep. Yep. Very famous. Well, thank you so much. It's been so awesome going deep with you on this conversation. The intent we have is super important. And as tastemakers, we have to align that with the future we're trying to get to.

1:34:18And it's subjective to where we're trying to go and how we use these tools. But I love the conversation. Thank you so much for spending time with me. Much appreciated.

1:34:26Amelia Wattenberger:Thanks so much, Adam. This was a lot of fun.

1:34:32Well, that's it. This show's done. Thank you so much for your patience, you all. You know, I know change is challenging. And I know information and lack of information when change happens is kind of a bummer. And I get that. And I will tell you that I want to share everything with you. I want to share where I'm at. I want to share where I'm going. I want to share where the show is going. And I want you to rest assured that nothing is changing, but kind of everything is changing in a good way. This show is going to get better. This show is going to have more. And I'm going to diversify a bit, build some software, go behind the scenes a bit.

1:35:07I've got some plans, but I'm in the middle of re-architecting things. So be patient with me. Be patient with the process of this podcast. And just know that I have good will for you and your best interests at heart. Many of you out there, I call my dear friends. And I am so thankful for you listening all these years, your loyalty, and of course, your friendship. But this show today with Amelia was very cool. Talking about agents. It's kind of all the rage, as you know. it's everywhere we can't avoid it but this replatforming the the perspective that she brings the table is such a unique perspective in her design skills her intention on intent the product is paramount and so cool and so going back through her journey talking about intent and what they're doing with it was so cool for me and i enjoyed it and i hope you enjoyed it too big thank you to our friends again over at rwx our friends at work os our friends at nord layer and of course our friends and partners at fly.io.

1:36:11The Beats, as always, by Breitmaster Cylinder. Those fantastic beats. And thank you so much for listening. That's it. The show's done. We'll see you next week.

1:36:44Win it. Win it. It really whips the... King of... Yeah.