Meet Ada: My AI Software Factory on Claude Code, Explained

0:19 Welcome, and what this show is

NickHello and welcome to Not Replaced Yet. My name is Nick Bushkar, and I'm here with my brother Jesse. Say hello, Jesse. Not Replaced Yet is the show where Jesse and I explore and explain how we work every single day in the age of agents and AI. If this is your first time here, Jesse and I run an agency servicing clients across the globe. We learn relentlessly by listening to experts and by building every single day.

JesseHello, hello.

NickIncluding nights and weekends. It's literally nonstop. And this show is where we share what works for us.

0:54 Today: agents, agentic architecture, and an AI chief of staff

JesseToday we're gonna be talking about agents, agentic architecture, and how Nick is building his AI chief of staff that he has named Ada. We're gonna dig into some of the nerdier details as we progress, but Nick, let's start out. Can you give us a non-technical explanation of what, or maybe I should say who, Ada is?

1:12 In 30 seconds, what is Ada?

NickSure. So Ada is my AI chief of staff. And when I talk about Ada, yes, I use anthropomorphized pronouns. So she essentially helps me organize my work, she helps me execute on that work, and then she also helps me think through what is possible. And there's so many more pieces and parts to it, but that's kind of the core of who and what Ada is.

1:41 Why the name Ada? Lovelace and Babbage

JesseNick and I are both building kind of a complementary agentic architecture. So I've been along the ride with you, Nick, but there's a lot that I don't know also. And that's why I was excited to dig into this and come up with some questions for you. But first, while we get started, why the name Ada?

NickAda. For you other historians or nerds or just general enthusiasts, you might know that essentially the first computer programmer out there, or kind of, was a woman from the eighteen forties named Ada Lovelace. And she was friends with Charles Babbage, who had this idea of the analytical machine. She was the one who recognized that you could sort of generalize that purpose, basically coming up with and anticipating generalized computers a century before they became a thing. So she's famous for other reasons too, but that was kind of her main contribution.

JesseI'm a huge computer history nerd and I love the story and I think it's a great choice. It's helpful that it's short too, with prompting and with relaying.

NickI like that, yes. And also slightly like AI and Ada. I don't know, there's a ring to the sound of it too. It was several purposes, but yeah, I just like it.

JesseAda contains multitudes, right? That's the takeaway.

NickExactly. Yes, she does.

3:00 Is Ada an agent? Markdown files and a database

JesseSo we know kind of what Ada does in a general sense, what Ada is when we're talking about Ada. And Ada is an agent, is an agentic architecture, is in that space. Do you have a really clear definition of, technically speaking, Ada is an agent? Or how do you think of that?

NickYeah, that's a great question, because there is something. If you zoom out and think of it in its most simplistic form, it really is a series of markdown files, which is just text files, and a database. That is really what Ada is. Because I did not create a large language model, obviously.

Ada is a direction, is a definition, is a set of memories and a knowledge base that Claude, and I generally use Claude, that Claude is the intelligence layer that acts on those things. And so it's an extension of Claude. And I want to talk about this a little bit later because I think that's an important point. But yeah, it really is. Ada is generally just a collection of markdown files and a database.

JesseWith the intelligence flowing through it is the picture I have in my head when this visualizes. It's like water going down paths and splitting across, like the architect. Or Plinko. Plinko, best prices, right? Plinko. Audience gets that one. It's a Bob Barker cut. No Drew Carey here.

NickExactly. I like that. A very relevant example. That's right.

JesseNo, I think it's also really important, because you've got some really cool demos, some visuals, and I don't wanna overly nerd out at the front end here.

4:50 What is a software factory, and where does Ada sit?

JesseBut I think it's important for the audience to understand this concept of a software factory. And this is a term that is out there in the world. It's not something you or I came up with. But Ada plays a role in the workflow of the software factory, because largely what you're doing is software development. It's development going through the software development lifecycle, and Ada is aiding in that process.

So what is a software factory? Where does Ada sit in that stack?

NickYeah, absolutely. So there's a lot of conversation these days about a software factory. Some people are big proponents, some are not, but it's essentially a loop. And what that loop does is, you have a plan, and I think of the planning part as a part of the loop. You have a plan and then you figure out what it is that you're going to build, and you sort of beat that up. And Ada, that's where I start. I talk to Ada and I say, hey, I have this idea for an app or something that I want to build.

And she and I chat back and forth. Now, she has been prepped to grill me in certain ways, to investigate and interrogate and adversarially push back on certain ideas and thoughts, for me to plan through what I want to build out. Then once that plan is built out, I get a brief. That's a document. That document then gets taken to another part of what I have inside my software factory, where it's the architect layer, and that architect layer takes that briefing document and makes it very specific. It creates documents that are laid out and organized in very specific ways. So I give it that brief, and that's basically another conversation between Ada and I.

And then finally I give that document, the architect document, to my software factory. I call it Archon. Archon takes that document and then starts a loop of development. And that loop of development has different layers to it. So Archon is the orchestrator agent. And then it basically goes through development and quality assurance and user interface, et cetera, et cetera, and all these things. And to your point, I do have this visualization, and this is kind of a part of that visualization. So there's something there too that we can get into in a bit.

JesseAnd I don't want you to talk about all the layers, 'cause those are future episodes.

That's what I was gonna ask, because if someone just listened to the highest possible layer of your architecture, because it goes a mile deep on all those layers. If you wanna share that first visual, unless you had a different setup for it, I think it's...

NickNo, let's do that.

JesseI think it's impressive to kind of connect that, okay, there's a lot of layers, there's a lot of this stuff going on. What does that actually look like? 'Cause what you made is super cool.

And while Nick's pulling that up, I think I come at this from a perspective of, the agentic architecture I'm building isn't a software factory per se. It builds software sometimes. But I'm gonna shut up because I just think this is so cool. Tell us, what am I looking at here?

7:51 The 3D model of Ada

NickSo, yeah, absolutely. First I will say that I think of the software factory as a subset of Ada. Ada is the overarching umbrella. And so that's my mental model of it. But we'll look at that in a second. So this, what we see right here, is this 3D model that I actually asked Ada to create of herself. So I opened up...

JesseGot it.

8:40 Inside the brain: memories, working files, database

NickI opened up my sessions and I talked to Ada and I said we were doing this episode and I thought that this would be a really good representation to explain what's going on. So there's a lot going on here. I can grab it and I can zoom it around, zoom in, zoom out. But I'm gonna focus in. I can show you here the core of Ada. If you could see this, this is the brain, right? So other things have faded out. And what the brain is, is sort of three separate things.

You have all of these are the memories. If you could see this here, there is a memory and there's one fact per page. If we can zoom in. And it says there's 160 of them. I didn't realize there's about 160. That's interesting to me. Because that's definitely a layer of this that I'm not, like, every single day keeping up to date on. I don't know where that's at.

And then in the middle here, right here, what we have are the working documents, the working files. That are essentially the files that have been pulled based on any given task. When I start talking to Ada, then she learns, or she's like, let me go look through my files and documents and see what might be relevant to this. So that's a part of it.

JesseSo the brain, and just to clarify, that's different per context window. If you're in a different session, that brain has pulled different documents based on what you're getting at in that session.

NickCorrect. That's correct. And then the sort of the lowest part here is a database. That's on Supabase, it's a Postgres database. And that has a bunch of different rows of data, of information and memories and things like that inside of it.

JesseAnd I think of the database layer, in my mental model I think of it as the persistent thing, right? Because it is shared across device. Long term memory is not the best analogy, but it's kind of effective.

NickIn a way, yeah. That's a good mental model. It's kind of a mixture of things across all of these parts. But yes, it's like memory and the brain are long term knowledge, but they work differently. And so anyways, so then what you have...

JesseWell, and it's a lot cheaper to write a fact in a markdown file than it is to make a network call and persist it with a third party service. So there's a hierarchy there too.

10:34 Inbound: email, phone calls, transcripts

NickTotally. Absolutely. So then here I just highlighted the next section, which is inbound information. So these are inbound things that are floating into the brain. And what we have is email, phone calls, transcripts, ideas, things like that. Just information that can be put into the brain.

JesseAda has an email address, I know. Does Ada have a phone number?

NickNot yet. That is definitely coming, but I have not built that layer out yet.

JesseI was gonna say, you could look at my Twilio work I was doing two weeks ago.

NickAbsolutely. So then from there, what we can see is we can look at the different agents. And there we go, they're highlighted there.

11:22 So what if Ada has an email address?

JesseHold on, can I actually ask a question, because I think it's interesting in the context.

NickPlease.

JesseYou just said, we just said Ada has an email address, right? What's an example of why that matters to you? You're working, Ada has an email address. So what?

NickYeah, absolutely. So there's many things. One obvious example is I get an email from one of our clients that wants an update or requests something that needs to get done to a site that I'm working on, and I forward that directly to Ada. So then the next time I go talk to her, I can say, hey, let's talk about that email. Let's talk about this feature that we need to build out.

So it's a very smooth way of keeping her up to date with what's going on. I have not automated... some people, and I think you might have done this, but some people like to just give full email access to their chief of staff or agent or whatever. I haven't done that yet for a few different reasons, but I think I'm going to eventually. But I like to keep it sort of leaner, and the information that she has access to.

So yeah, that's one. Another one is, we work with a bunch of software, and those different apps give you different warnings, like security warnings or things. And what I have across those is, many of them are automatically set up to forward directly to her. So then she can know that for this app or this whatever, we need to look at this security issue or something. And so it's just very, very applicable, very relevant.

JesseSo Ada having an email address is just one inbound port for you to stay super organized, way more organized than without it. 'Cause it would live in a threaded inbox that we've all grown to hate, like managing your inbox. No, sorry, cool. Keep going. Then the agents come in.

13:19 Agent sessions and the factory loop

NickNo, absolutely. So let me get this highlighted back up. So the agents, the way that it's separated here, which we see, is there's agent sessions. Workers, one per job, per company, whatever. And then what we were talking about earlier up here is the factory loop. Brief, plan, build, merge.

So what I think of this is, agent sessions is at any given time I have a ton of different Claude Code sessions up. And those Claude Code sessions, oops... may or may not be, most of the time they are, but here, I'll just drag this over because I actually have it open right now. So if you can see this, this is my terminal multiplier. These are three different agent sessions with Claude that I've been working on, or that I was working on earlier today. And this is me doing one-on-one work, interacting with three in three separate sessions. And that's what this larger ring represents, is just me individually, focused work, talking to Ada about different things at different times.

So to the factory loop that we were talking about, that's what this ring over here represents. Once a brief has been created, it pops into this circle, right? You go from brief to plan to build to merge. And there are more parts to that. That is sort of a very unnuanced way of looking at it. But it's like, you create a plan, and then you set that plan off in a loop and it gets executed on. And then sometimes I put that in auto mode. Sometimes I put that in checkpoint mode, where then it escalates important decisions that I need to be aware of, that I need to have a say in. And there are different things going on there.

15:08 Focused work vs. background work

NickBut to this point, I do think that something is important here, is that one of the things that Ada really helps me do is she helps me, I said, organize my workflow, my workload. And so I have this, I think of it as my focused work and the background parallel work that I'm doing. So I'm focusing on one thing. Well, I might be focusing on a few things, because I can have a few different conversations going on at once with Ada. And in the background, I've set off three different factory loops. So she's working on three different problems in the background while I'm talking to her in three different sessions in a focused manner.

And so that helps me do a lot more. It helps me get a lot more done.

15:49 One shot, under an hour, no PowerPoint

JesseI really like the model of focused and background work that you're managing. These visualizations are so cool. Yours looks a lot better than mine. As I'm building my system, I've built a visualization or two, and this is just way better. And I would blame it on model intelligence, because I think I made something maybe earlier than this version that I'm comparing it to.

But I think I used Fable 5 to build the last version of it, and yours still looks way better.

NickI don't know. This is a one shot with Ada, talking about this episode, saying, I wanna show you off. Let's create a 3D representation of who you are. And it was Fable. And so I don't know, this was just a one shot.

JesseThat's a good point. I think a lot of people who don't build like this, and who haven't had the front row seat, might... the naysayer amongst the crowd is like, why would you waste your time? This is cool, but you're not... I mean, correct me if I'm wrong, you're not using this visualization for any planning, right? You made this for the show. Right, exactly. This is a representation.

NickNo, I literally made this today in the background while I was doing other stuff for this show.

JesseA visual. This comes to one of the things you've been saying during day job hours about the death of PowerPoint, and kill PowerPoint. It's like, look, look, look what was made in a very short time with the appropriate tooling, with the appropriate architecture, compared to a PowerPoint.

NickAbsolutely. Under an hour, as a sub agent inside a Claude Code session, using Ada.

JesseWhich was non blocking to the rest of your work.

17:22 The split: the board, your focus, the background

NickExactly. Well, so to move on, and it just kind of expands and extends the last point that I was making. The next layer here that I have is called the split. And if you can see at the top, it says the board, Nick. And this is some AI writing, but he decides, Ada recommends. So there's the board, there's needs your focus, and then down here is that background work, this green circle at the bottom.

So it's the split of the work that's going on, where I have focused work, things that only I can do, things that absolutely need my attention. And I have... it depends on what I'm doing. It can be one single thing, or it can be, like I said, multiple sessions where I can just sort of jump back and forth between them. And then in the background, my goal, and I'm not perfect at this, but my goal is to always have something running in the background, at least one thing going on, if not many things.

So that's the split. And the board, it's called the board because, like I said, Ada is my chief of staff. It's the idea of the board of directors. But it's, I don't know, it's unimportant as I see it.

18:36 Side quest: Inside Out

JesseYou know what this makes me think of here? This is a complete diagonal side quest, but if you go back to the last visualization, have you ever seen the movie Inside Out?

NickYou know, I never have. I've heard it's great.

JesseYou never have? You're missing out. You gotta watch. It's great. And the sequel. There's a model of the brain. The reason it makes me think of that is that in Inside Out, they are basically having an opinionated personification of the conscious experience through these characters. And then also not just the characters, the model they use where they go to long-term memory storage and they do this other thing, the imagination. And so it's fun, it's a kid's movie, but it is also thoughtful about that. And this, I don't know, it rhymes with it in a much nerdier way.

NickThis is Ada inside out.

JesseAnd it's like Amy Poehler's there shaking her fist or something. Anyway, you should watch the movie.

Okay, so let me take a milestone here. We've now looked at the dissection of the architecture of Ada through one lens. You could represent Ada in so many different ways and this is so cool.

19:45 What has Ada not solved?

JesseAnd it helps you stay organized with communication. It allows you to have a double attention span, a kind of clunky way to say it, right? You have a two sided attention span, which is background work and focused work. Which I get, there's kind of this flow state about going between different contexts and jumping back and forth that is enjoyable. I think we both really enjoy that. And so it's helping you in all these ways, foundational and specific and everything.

What has Ada not solved for you? Is there anything that you thought was gonna be the perfect thing and you're just not finding the value in the current approach? Is there anything on that list?

NickYeah, there might be a few things on there. The first thing that popped into my head, this is kind of in that direction, is I initially set Ada up to talk to me through Telegram. Telegram is a chat app, right? And there's an interesting point, there's definitely parallels to OpenClaw and Hermes and everything, and we could talk about that maybe in a little bit.

JesseLike OpenClaw, is the... yeah.

NickBut I initially set her up to be able for me to be able to talk to her through Telegram, and I just never really got into that mode. I have different surfaces that I do talk to her on. It's mostly my laptop. But Ada lives, and this is a whole other thing, Ada lives on Aevum, which is her own dedicated computer that I have here at my house. But that's a whole other episode, really. We could talk a little bit about that if you want.

But so the Telegram thing didn't really work. And one of those things was push notifications, or push messaging, to check in for things that were not easily inserted into her memory. And what I mean by that is, okay, if I forward her an email, or if I have a direct chat with her, then she knows about those things, and whatever's important can go directly into her knowledge base. But if something happens off the board... now, you and I, this is another thing, we've set up our Zoom calls so that our transcripts automatically get downloaded and then Ada can read those transcripts. So we have that set up. But say you and I are having a conversation that doesn't get recorded, or whatever it may be. Now, the concept was I want Ada to check in with me, to talk to me about things that happen off the board, outside of her scope of awareness. And it's just not been a very useful thing.

And I think maybe partially it's because I've built my life to not really need to do that, honestly. But that was one of the first things that came to mind. I know there are other things and other experiments that I've put forward that haven't worked. Honestly, they haven't worked and I've forgotten about them. That's a whole other layer of it.

JesseOr we find a different way to get the value out of it, right? You could kind of reframe it, because there's so many ways to solve a problem in this space.

22:56 HQ and the session monitor

NickYeah, absolutely. You know what, I have an interesting one, and it actually takes us to another visualization. So if you'll let me take this for a second.

JessePlease.

NickOkay, so this is just a visualization. Now, what I do use is something that I call HQ. And I am still building HQ out. And HQ has many different layers and feature sets and things, but I'm starting to use it more and more.

23:24 Building the email inbox UI in the background

NickAnd so this is a very basic view of HQ. Now, this is the session monitor inside HQ. This is a monitor that tracks every single Claude Code and Codex session that I run. And it tracks everything that I'm doing. And these are two active sessions that have been running the entire time while we've been talking here. So things going on in the background, like I said.

So if I click into this, this is actually building out something for HQ right now. This session is building out my user interface for Ada's email inbox. Because right now I email something to her and then it's kind of a black box unless I ask her about it. I wanted a user interface for it. I wanted a UI for that email. So it's actually literally being built right now. And if I click into this, you can see, here's every single thing that she's doing. There's a sub agent handback, there's a message, there's a session lifecycle, a tool call end, and it's tracking and storing everything. Of every single session that I've been running for months now, six months at least. And so this is useful for many different reasons.

24:28 Village view: more fun than useful

NickAnd then this is something that hasn't been useful, but I like it because I'm nerdy. And if I go up here and I click on village view, I've created this nerdy view of all my agents and how this works. So this is the village view, this is the town. And each one of these little... you and I love some Nintendo and Super Nintendo RPGs. And this is, of course, if you also are a fan of those things.

JesseHey, shout out Adventures of Elliot. We're just finishing that game and that's a new one. But yeah, Final Fantasy, all of them.

NickI don't know about that game, but okay. Cool. I'm stuck at Final Fantasy three, or six for those nerds.

JesseIt's good. You gotta play. Six, right? Get it right.

NickBut exactly. So each one of the characters is an actual representation of one of my agents that's running. So if you hover over here, I don't know if you can see it on the shared screen, but it says HQ inbox, and I click on this and it brings up the little chat window, just like if you were chatting with someone in the town, and it tells you some information and so on and so forth.

JesseIt's pretty small, but yeah.

NickYou can go down here and these are agents that have been done. This is what I mean. I haven't really used this. I've found it more fun than useful. And so I do want to continue to build this out. And I have my dog Casper running around the town right here. And so maybe I'll use this. A mutual friend of ours gave me the idea of, especially one day, I could just have this up and running on a screen in my office, almost as art. And that scratches my nerd itch too. So there's something there too.

JesseYeah, no, I love it. We both did the visualizations early, and I went galactic planetary solar system. Again, not as cool as this. And you went total 16-bit RPG. And I thought it was fun, but then when I would ask, hey, are you using it? It's like, I want to, but I don't really find a way. I wonder if that's the lossiness of the UI. It's like there's just so much to manage in a real, fast paced agentic architecture. Do you feel like it's that, or maybe it's missing one ingredient?

26:52 When a visualization earns its keep (5,000 agents)

NickHere's the best thing that I've come up with so far, and I don't know if it's quite right, but I think a visualization like this could be really useful if you are using a massive amount of agents. Because what you need is the visual of what's going on, because the visual is something that is an instant indicator that something's happening. Rather than, if you, like you and I, we listen to the Moonshots pod and Dave Blundin talks about spinning up a thousand agents, five thousand agents, whatever.

So if you have five thousand agents running, no ledger of text is gonna be useful to you, in my opinion, at least at just a glance. So a visual like this could be useful in that, if you see characters huddled together and it looks like something, then it's doing something, and you have this idea of, okay good, that's my developer agents that are working over there in the corner of the town or something. So maybe it's something like that. But in the end it might just be a nerdy fun thing. I don't know.

JesseThat's a great... yeah. Fun's okay too. I could see what you're saying. It's almost, zoom out and take a civilization approach, and then you have scale to play with in your visualizations that matches the scalability. I think that could be cool.

NickYes. Yeah, sure.

JesseThere are a hundred different rabbit holes in this process that we could go down.

28:19 The better hour spent

JesseI think our mission was accomplished. We wanted to give people an introduction to Ada and agentic architecture. I would say, to the audience, if you are interested in this stuff, and maybe Nick, this could be my final question to you. I'm leading the witness, certainly, but I find my work in my system is, every hour I spend using it is a better hour spent. That sound bite rings. If I were to have done an hour of work in the before times, it just was not as efficient, not as productive, not as thoughtful, not as creative. Not just efficiency. You've got this idea of depth and width of bandwidth. Do you agree in the better hour spent piece? And how do you think of it giving you that better?

29:14 Force multiplier: width, not just speed

NickYes, I totally agree. So Ada has absolutely made me faster. I think I've even made the point while we've been talking. If she's helping me think through things, then I am doing things at a quicker pace. But the bigger gain, absolutely, she is an absolute force multiplier. And I will say, of course, that Ada is built on top of Claude, and that's the foundational thing. That's the real new layer, the new foundational piece of technology that is enabling this. And Ada is, I think, a piece of technology on top of that that has helped me even more.

JesseIt's like the force multiplier, yeah.

NickAnd so I would say the bigger gain is what I think of as width, right? So I have overnight lanes that produce drafts, and she researches things for me, creates plans. I do a security investigation and look for incidents and things like that. And so it's just so many more things are going on all at once that I have. It's the force multiplier. I cannot evangelize this method enough.

30:19 Start small, markdown brick by markdown brick

JesseAnd even a small start. I think maybe this is the wrap up. Anyone who's thinking about starting this and seeing Nick's massive 3D brain, from Minority Report, as I date myself, don't feel intimidated, because everyone's agentic architecture matches to them. And I'd like to do future episodes on other approaches to this, 'cause we're building from scratch agents. But I think that even day one with my system, you with your system, it was still an improvement, 'cause it's just a little bit towards something bigger.

NickI completely agree. And yes, to anyone who sees this and thinks it's large and daunting, it makes sense. But it literally is a step by step process. What you do is you start with one piece of functionality, and then when that works for you, you move on to the next thing. And so this started way smaller, and I've been building this out for months and months now. And so that's why it's larger right now.

JesseMarkdown brick by markdown brick.

31:33 Outro

JesseAwesome. Well, thanks everyone who made it this far. We really appreciate it. Nick and I make Not Replaced Yet in our free time. As Nick said, we're very busy, and very busy learning, and very busy doing our job. But we love this, we love sharing stuff. So if you're here, please give us a like, follow, subscribe. We really appreciate that. And otherwise, we will see you next time.