# The next level of AI engineering — Kitze M — session 2026-09-23T16:15:00.000Z → 2026-09-23T16:40:00.000Z _101 transcript lines · 61 slides · source: full recording_ ## Transcript **[00:00:00 · 0]** Right. Slide. Boom. You guys can see it. Perfect. **[00:00:03 · 0]** I just realized how bad it sounds. We call it Master Tink. I just like tinkering with stuff and we form this entire thing that I'm gonna tell you in a second. Are we gonna stop the music or am I gonna have a background soundtrack? Okay. **[00:00:15 · 0]** It's going away. Cool. So, I'm the kids on X. If you know me on X, we probably disagree about many things and we probably argued about many things. So, I have twenty something minutes before I need to catch an Uber. **[00:00:26 · 0]** That's gonna be fun. I just need to get my cursor over here so I can click and it will be perfect to start. So, usually I plug in my 39 products that I'm working on. This time I won't. I'll just have one slide. **[00:00:37 · 0]** If you are someone who tinkers with, like, local LMS and hardware and self hosting and a bunch of other things, we have a group of crazy people called tinker.club. That's the URL. And basically, every person within this club is a copy and paste version of themselves. It's hilarious to see that everyone has the same interest from, like, hardware and three d printing and AI and SAS and business owners. If you wanna join us, it's at tinker.club. **[00:00:59 · 0]** So previously at AI Engineer, I've given a talk, called from Vibe Coding to Vibe Engineering, which still many people are going to that transition. So if I would do the talk right now, it would still be relevant because so many people are just moving from Vibe coding to Vibe engineering. But I can already sense the shift within me, within my own apps, and within other people's. Like, I can see what's... Everything that's going on on Twitter. **[00:01:20 · 0]** And we're slowly starting to move to the next level of your engineer. Nearing, you probably heard a term like factories and software factories and slopware factories that I call them. And you can see that within q four of this year and within next year, it's gonna be more and more and more about going from like Vibe Coding and even this old version of Vibe Engineering to something new. But before we talk about something new, we gotta go a little bit back. So I'll take you just a few slides through what we went through in the last few years. **[00:01:48 · 0]** There's still many people who are using Copilot and they're pressing tab and they call that AI, but we went through that for the longest time. At some point, we were copy and pasting from chat gpt. And if you remember, if you're old enough, like, few years ago, it couldn't finish a function, and you have to beg it and be like, could you please finish that? Then you wait for the next one, and then you copy and paste it somewhere. Then we had to beg again. **[00:02:07 · 0]** But that time, we were begging for JSON. We're like, can you just not put markdown in it? Just I need clean JSON. And it'll be like, sure. Here's some markdown and some JSON. **[00:02:16 · 0]** So whoever started with AI now, you're very, very spoiled. Then we, we focused on editing one thing. So there was no agents, but we would just use AI, and it would change a file. And we're like, oh, this is dangerous. It might do something. **[00:02:27 · 0]** Right? And then very quickly, it felt like we went through agents, but we were still careful and we were just reviewing every permission. We're like, be careful. Don't touch the the thing. And then, like, literally one year later. **[00:02:37 · 0]** Like, when it asks me if I can run this, like, how the fuck should I know? Right? You just read the thing. You're like, yeah. I guess. **[00:02:45 · 0]** I I fucking hope so. Right? So now everyone is in this full YOLO mode. There's not a single person that I know that's not actually running on full permissions. Like, take my production databases, customer data, GDPR, whatever, man. **[00:02:58 · 0]** Take everything. Just finish the the job. So we ended up here. Basically, we have a few categories of people that are working with AI right now. We have the slop grenade throwers, which is most of us. **[00:03:08 · 0]** We're trying not to be, but we're just throwing the slop grenades. We have the meat proxies. Have you heard this term? Like, this is one of my favorite terms I've heard lately. Like, I absolutely love it. **[00:03:15 · 0]** And we have actually engineers, which are, like, not a lot of people, but we're trying to make, like, everyone an AI engineer, basically. So slop grenade throwers, like, everyone is kinda doing this, that because of the speed of AI, it's very easy to just prompt an agent, and then it generates a bunch of slop. And then you don't read it, and you just throw it at your colleague. You're like, good luck. Good luck. **[00:03:34 · 0]** Hopefully, it doesn't explode production. You figure it out. And then we have, like, the meet proxy, which is, like, if you don't know what this is, it means you're sitting between your manager or your supervisor, whoever it is, and they ask you to do something. And then you literally take their prompt. You put it into an LLM. **[00:03:48 · 0]** Then you take the result from that LLM. You barely review it. You give it back to the person. So you're just serving as a meat proxy. And currently, there's so many companies, and this would have to change, that they're actually consisted of layers and layers and layers of just meat proxies talking to each other. **[00:04:02 · 0]** And I think next year is gonna be brutal because many people, they're gonna wake up to the fact, they're gonna go to the mirror and be like, oh, fuck, and they're gonna get fired. Like, it's inevitable. I've been talking in my talks about this, and it's gonna happen, I think, around next year. The actual engineers actually try to build systems around their processes, and no one has the right system. So this talk is about, like, my road there. **[00:04:21 · 0]** I can see, like, two camps of people. It's always been a division. So we have the desktop people, and we have the CLI people, the hackers, the thinkers that they'd like. I always hated the terminal, by the way. I always hated CLI. **[00:04:31 · 0]** I was a big hater. Like, give me a beautiful UI that I can click on. I always hated this tinkering type of stuff. What's mind blowing to me that we had like three years of orchestrators being released from cursor to conductor to everything else. Everything. **[00:04:45 · 0]** And like no one is even trying. It's just a copy and paste of the same layout. On the left, you have a sidebar with chats. In the middle, you have chats. And on the right side, you might get a git diff. **[00:04:54 · 0]** And everyone is like, we just raised 37,000,000 for a new orchestrator and the screenshot. We're like, is that codex? Is it cursory? It's the same shit always, and no one is even trying. I'm gonna try to show you mine eventually at the end of the talk, so we'll see if you like that. **[00:05:08 · 0]** CLI camper, the vanilla ones, and I'm going between back and forth, like, I launch a desktop app and then I go back to the CLI. And then we have, like, the absolute crazies where, like, if you haven't heard of the term oh my pie, don't even research what is oh my pie. But we have people who have these Oh My Pie instead of Herder, instead of Omerchi, instead of VM, instead of Proxmox, and ask them what they shipped, and they'd be like, yeah. But my factory works, I guess. So I was the one roasting all of this. **[00:05:33 · 0]** I always hated it. Like, self hosters, like, I was always the one. Just pay Vercel, pay, railway, pay whoever. Don't bother with this. And then I just full went onto the other side with Tinker Club. **[00:05:41 · 0]** Because the more people post about this, the more I wanna do it too. So I have, like, Linux boxes and tail scale and a bunch of bullshit, basically. So we have tools versus methods. We're going to talk about methods. So so we have, like, the two tools, like CLIs and desktops, and then we have, methods. **[00:05:55 · 0]** So the one way... Like, I've been going through all of these, and then I'm going to tell you which one that I landed on. One chat and problem until work. This is like, I would stop opening multiple chats for different tasks and I would just keep one. And especially with the smart compaction of codecs, I can just dump a bunch of things in the queue, and I don't even know where it compacts. **[00:06:12 · 0]** I think it doesn't even tell you that it's compacting. So, it would just work and pick up the next task and pick up the next task and the next one. And it would work and ship, but in between every task, there's no, like, management where it's, like, launches a sub agent to review and then another one to test to just hope that it's gonna go through all the tasks and it's gonna finish all the tasks. Kinda like it, kinda dislike it. Then we have, this is actually one chat per task where you have like a folder in codex or whatever. **[00:06:36 · 0]** And then for every task that you need to do, you start a new chat. But if you don't use work trees, you either have to guess whether the features are gonna overlap each other or you have to use work trees and then you need a chat on top of those chats to do something with the work trees. So it's kinda getting messy. Then we have, like, work trees and hope. So no... **[00:06:55 · 0]** Nobody knew about work trees, by the way, until, like, seven weeks ago. So shut up. So now everyone's using work trees, and everyone is a Git engineer, and you know stuff. Right? But the problem is these are still not fully isolated systems. **[00:07:05 · 0]** They're a different git thingy, but they can still nuke your database equally as not using a git work tree. And then we have all of these terms, which a lot of times I hate because they're overwhelming to learn all of this. Have you heard of gauntlet loops? First, we had loops and everyone was like, what? And now just like, oh, I just did this using a gauntlet loop. **[00:07:25 · 0]** Like, I keep... We keep inventing all of these terms and some of them are useful, some of them are not useful. So, yeah. And finally, we're talking about software factories. If you ask any of your bots to make a research of all the articles published about software factories, your mind's gonna be blown, and you're gonna read through half of it and be like, oh, just let me prompt for fuck's sake. **[00:07:41 · 0]** Let me be a mute proxy. Because it's very overwhelming. Because, if you look at... If you zoom out, like, almost nobody has an actual software factory and people are just scraping it together, and we don't agree on what a software factory should be. Should it be on a level of a company organization? **[00:07:55 · 0]** Should I have, as an engineer, a software factory? And we end up with this slop grenade fight. So if you look at any GitHub repo, any organization, you just have someone who opened a PR using an LLM. You have a beautiful read me with, like, emojis and paragraphs. We never even bothered to make something bold on GitHub, but now we have, like, paragraphs and dividers and everything. **[00:08:15 · 0]** And then their colleague responds with even more beautiful review. It's even longer. They're paying attention. Suddenly, things you're like, you didn't write that, you motherfucker. You know it's by an LLM. **[00:08:25 · 0]** So now we're going between, like, two meet proxies serving two LLMs, and they're talking to each other. And nobody realize... And eventually, the task has been submitted either by a customer or by a superior. Right? And no one is realizing what what what are we doing here? **[00:08:38 · 0]** We need to, like, untangle this so it's actually predictable in some way to know what are we we doing. So I throw one back. My AI reviews your AI and blah blah blah. Eventually, they meet in the middle, but we need to find a solution, and we need to make the process to evolve. So my road to this from terminal to desktop apps terminal again. **[00:08:56 · 0]** I kinda recently went back to the, CLI. Who has heard of this? Heard of this? Get it? Heard it? **[00:09:02 · 0]** Get it? You get it? Okay. So if you don't know what this is, it's basically like a t max thingy that if you do remote coding on any of your devices, the sessions are not gonna die. And in the terminal, they're gonna be very nicely organized so you don't lose track of of what you're doing. **[00:09:16 · 0]** And, ironically, like, I've been going between so many setups. I even made my own orchestrators and stuff. And, like, last week or two weeks ago, I just set up Herder. We did, like, the simplest terminal, and I shipped the most. And I'm like, maybe we were overcomplicating the entire software factory thing, but it really depends. **[00:09:31 · 0]** Are you working with people? How many people? How many teams? Or are you solo or shipping shipping what exactly? I moved to pi. **[00:09:39 · 0]** Pi rocks. I like pi. I was making also fun of everyone who used pi. I was like, just be a person. Use Claude. **[00:09:44 · 0]** Use codex. Use whatever. And then as my quotas were going, you know, like, my Claude quota would die, and then I would switch to codec CLI or codecs desk or whatever, they have completely different behaviors, like how they launch sub agents, how they do compaction, how they do this, how they do that. So I would switch between the cursor CLI, between cloud CLI, between codec CLI. And as I switch between them, they completely have different behaviors. **[00:10:04 · 0]** And this was driving me crazy. So when I build an orchestrator, I build it on top of the three. But it's so hard to build an orchestrator that needs to serve... Like, it's not only the CLI. It's like they have different methods of doing things. **[00:10:15 · 0]** And with Py, you you just have a CLI that does nothing out of the box. So if you want it to compact, you need to install a plug in. If you want it to launch sub agents, you need to install a plug in, and you fully control the process. So it's like a Linux of harnesses, which I see none of you are like, oh, yeah. Fuck, yeah. **[00:10:30 · 0]** I would love to do more Linux of things. Right? So Herder also has this nice feature. I think they added it recently that if you have Herder on all of your machines, when you open it on any of the machines, you're gonna see your, I I don't know, MacBook Pro, Mac Studio, Linux, VPS, whatever, and it's all in one place. It's kinda like very nice. **[00:10:47 · 0]** I shifted this year to using my laptop only as a remote control. So I started by the cloud and whatever, and now I'm moving to my own things. But it's this laptop or my phone are only remote controls for things that are happening in the cloud. So I went into this deep rabbit hole of, like, I don't want my code because I spilled milkshake on this laptop, and I had, like, an unforced GitHub repo. And thankfully, they saved it, but, like, I thought it's gonna die. **[00:11:08 · 0]** And, like, I don't know, like, a month of work would have died with this laptop. And I'm like, I need to move all of my work somewhere and just use any laptop. I even use my ten year old MacBook Air to just remote into my setup and do things remotely. So So I started first by shoving things in a Mac Studio, but I realized that Mac and macOS is not made for remote agents. So then, like, probably after fifteen years of not even thinking about this, I was like, fine. **[00:11:31 · 0]** I'll just try and maybe Linux will be the good thing. So I started renting a VPS for, like, $128... Euros per month, and I was like, if it makes me money, it doesn't matter, whatever. Then I did the calculation. I'm like, but if I pay this for two years, I can just buy a Linux machine. **[00:11:43 · 0]** And suddenly, you have a Linux in your basement. But I realized I have a gaming PC, and I don't like Windows, and I don't like gaming because this is way more fun than gaming nowadays. I don't wanna game. I don't wanna watch shows. I just wanna tinker with the with the agents. **[00:11:54 · 0]** So I just decided I'm gonna format this PC. I'm gonna put disgusting Linux on it. Right? But Windows is, like, up there. And then I realized that this works. **[00:12:02 · 0]** It's great, but I don't have enough RAM. So I started searching for part. Like, I went so far away from actually shipping value to my customers, and the rabbit hole goes even deeper. So I was like, this is 32. I need to get more, but I need to upgrade the motherboard, but I need this, but I need that. **[00:12:16 · 0]** I ended up with a second Linux PC. So I built a second Linux PC from scratch. And then someone was like, why do you choose which Linux to do... Distro to install? You can install Proxmox. **[00:12:24 · 0]** Now who is nerdy enough here to have set up? Because I'm going to skip this. Okay. We have a bunch of hands. So, this is a really cool thing that I used to look at all of these things we discussed because I had to go into the configs and click thingies. **[00:12:38 · 0]** Now, anything that I mentioned here, just assume I'm not smart enough to deal with this. I just told my agents, hey, my friend told me to use Proxmox, use it. And then boom, it's getting set up on the thing. So this is nice because actually on one computer, like let's say you have a PC, you can have... This is like a higher level OS. **[00:12:55 · 0]** You can have multiple Linuxes. You can have a Windows. You can have Windows XP. All of them as like virtual machines. And then your agents do everything beautifully. **[00:13:02 · 0]** You don't even need to manually tinker with anything. They also have these things like VMs and containers. So basically, you can like self host on your own computer a bunch of things that are available on your network. And you can actually run full VMs. So my agents, like my Hermes agent and whatever, I've given them small VMs so they cannot break out. **[00:13:19 · 0]** They cannot do something stupid. They just have, like, mini computers within my computer. So how I proximoxified, I just plugged in a KVM, and I went to codecs, and I was like, please finish this. I don't want to look at it. And then it worked, it worked, it worked, and then I saw it. **[00:13:32 · 0]** I was like, cool. It works, and I don't want to, touch it. So a lot of times, people start remote coding, and I see, like, people who send me screenshots. They have some weird IP addresses and shit. So you can do, like, a very, very nice setup. **[00:13:43 · 0]** If you're remote coding, you can just tell your agent to set up, however you pronounce that, and p m two. And then all of your processes are gonna be, like, magically managed and on nice URLs. You can have your blog, whatever your domain is. And as subdomains, you can have, like, dev processes. So anytime you wanna see how your landing page looks like in dev, you can go to my beautiful landing page -dev..myurl. **[00:14:08 · 0]** Whatever, and you're gonna see it, and you're gonna remember it, and you're gonna bookmark it. It's not gonna be a random IP address. So that's like a nice, tip there. Agent sets it all up. So the entire point of this, people ask me, what's your config? **[00:14:18 · 0]** And I'm like, how fucking know? I haven't seen a config in two years. I just wished for things and the things magically happen. Then I hit the limits because, of course, all of us are dealing with this. I have a tip here. **[00:14:28 · 0]** I'm paying for four $200 subscriptions. It's still cheaper to... Than the humans. It's, like, kind of fine. But you need to manage them in a different time. **[00:14:37 · 0]** So there's this thing you called codex load balancer, which you can, like... What? You got... You guys don't use a load balancer? I thought this is gonna be, like, widely accepted, and people are gonna be like, yeah. **[00:14:48 · 0]** Yeah. I do that. So this is basically, like, one endpoint, but when you hit it, it chooses from which one to siphon from. So it knows your limits, your weekly limits, your daily limits, whatever, and it just like my limits never expire basically until they released Astra, and then they were gone in a day. So with Astra, it was like a motherfucker. **[00:15:04 · 0]** At bonus, you can do live voice on the same endpoint, and you can generate images on the same endpoint. So you can give this endpoint to all of your apps, and they just use the load balancer. And it gets worse, guys. Then Then you have, like, that's for codecs. But what about the other ones? **[00:15:18 · 0]** I have, like, Cursor, Gemini, and a bunch of other things hidden behind this easy CLI proxy API, which is like another load balancer proxy. And then on top, we have an AI router, which is like the master endpoint. So when you... When I wanna code, I talk to it, and then it figures out who to ping from all of these things. And I still haven't shipped much. **[00:15:36 · 0]** So as I'm going through the slides, just be like, you can be more productive with just like pi, herder, and and whatever. So every model, one endpoint everywhere. It doesn't, create new capacity. Just trying to squeeze out from a little bit of cloud, a little bit of code exist in that to actually make it work. So my connections were scattered in one problem here. **[00:15:53 · 0]** And, like, this would be one slide that I take a picture of just so you remember the name executor if you haven't set up executor. I've been spreading the word around more than I did for OpenClaw because this is the shit. Like, it's absolutely the greatest because I would wanna switch setups. I would wanna try, I don't know, or this and that. And anytime I would have to connect my Gmail, connect my calendar, connect my GitHub, connect my CloudFlare, connect like all of the connectors. **[00:16:18 · 0]** And what this is is like a house of connectors, like with one MCP. So I've plugged in everything to my executor, and now if I wanna try muse by by Meta, I can just tell it to use one MCP. That's my executor. And through the executor, it's going to ping all of my other services. So it makes it super easy to switch between, I don't know, pi, clod, codex, any of the bots. **[00:16:36 · 0]** It's really nice. Then I realized this is open source. It's on my GitHub. It's called Skillbox. I realized that we're doing it wrong, that our skills are stored in markdown files. **[00:16:46 · 0]** Like, I had, like, a GitHub script that synced my skills across spaces, and it was very messy. And I would, like, set up a new computer. I'm like, oh, I don't have my skills here. And I'm like, if we have executor for MCP, and executor actually holds all my MCP, why I don't have a single MCP for all of my skills? And then it turns out, as I was working on this, they made it an official spec. **[00:17:06 · 0]** So now in the MCP spec, you actually have something called skills over MCP. So instead of having 30,000 markdown files here and there, you can just self host your own... Like, you can tell your bot, hey. I like this skill box. Set it up for me. **[00:17:17 · 0]** It's gonna give you a nice URL with, like, a graphic interface and everything, and all of your skills are gonna be there. And not only that, there's a thing called skill bundles. So So you have, like, a personal agent, a coding agent, and other things. My personal agent needs, like, four skills here, and my coding agent might need another 20 skills, but they don't need to know about each other. So if I set up my open claw, Hermes, I'm gonna point to my personal skill bundle, and it's only gonna work with those. **[00:17:40 · 0]** And it's not gonna be like, oh, what is CloudFlare? So that was very annoying that all of my skills were for everyone and in one folder with a bunch of markdowns. So now my setup is portable. I can just take it and just plug into something else, and it works. I'm currently working on something new, which I'm not sure if I'm gonna finish. **[00:17:55 · 0]** It's like a layer on top because I realized, like, we have skills. We have MCP. And Codecs, the Codecs app calls the layer on top, they call it plugins. Right? So if you go to Codecs and you install, like, a MacOS development plugin, it's gonna install MCPs for you, and it's gonna install skills for you. **[00:18:10 · 0]** So I'm like, we need some new names, like, I don't know, like playbooks. And, like, I have a bunch of ideas. Maybe I'm gonna present that in the next presentation. You're gonna see it on my Twitter. But the idea is, like, I can see Skillbox and executor kind of merging, and a couple of layers on top would make it way nicer to pack your entire setup and just plug it somewhere else. **[00:18:26 · 0]** Like Cody by Ken C. Dodds, if you've seen Cody's, like, kinda that. It's hosted on CloudFlare workers and a bunch of complex terms that I don't actually understand. But his idea was basically to take his brain of his agents, put it into Cody, and then he can just port it anywhere he wants. So Cody is like a skill box and executor with a bunch of goodies on on top. **[00:18:45 · 0]** An executor is coming out with a new version, version two, which is gonna be similar to Cody. It's very hard to keep track of all of this, but I hope I'm gonna plant some seeds that gonna tell you where things are going with this. So our RAM, our our our brain actually cannot keep up with everything that we're doing. So we have power. We have the four subscriptions, we have a load balancer, we have a proxy, we have this and that. **[00:19:06 · 0]** And I decided that unless I build, like, my own orchestrator to organize all of this mess, this will never go anywhere. It's gonna be very messy. So my... From everything that I mentioned to you, like, how I was actually working through things, this is what I landed on right now on this day today. So I landed on realizing that one chat where it constantly compacts and I dump a bunch of tasks is not going anywhere because there's gonna be plenty of conflicts. **[00:19:32 · 0]** Then we have the other thing where I have multiple chats. They're gonna trample over each other. Then we have the third one with work trees. That's also problematic. So I realized there was this article, I hope it's in my slides here, that you get one chat, then one work tree, and one PR. **[00:19:47 · 0]** So this is what many people were doing. I personally, as a solo indie hacker, let's say, I wasn't even working with PRs. I would just ship everything on main and go in a cycle. But now I realize I wanna set it as a goal to have, like, a very high quality PR with all the type checks, with linting, and a bunch of things. And I want to basically automate the process there instead of doing things manually. **[00:20:06 · 0]** Because it happened, like, I started using only work trees. And one of my work trees, because it's just a git thing, it wiped the production database. Luckily, I had a backup. And it's like, oops, sorry. I was trying to see the database to test, but I'm not testing in a container. **[00:20:18 · 0]** I'm just testing on your machine, and it just basically wiped everything because a work tree is not an actual sandbox. I have the slide here, orbs. So this guy from amp code, Thorsten something, he was talking about orbs, but my brain was so overloaded by these new phrases. And I'm like, fuck you. Like, orbs are not happening. **[00:20:36 · 0]** And I just tweeted that a couple of times, like, orbs are not a thing. They're not gonna be in our dictionary. Shut up. Don't talk about orbs. And basically, his philosophy was that anytime your agent works on a task, you need to give it like a mini sandbox, mini VM, mini computer thingy where it can work in it. **[00:20:51 · 0]** And as it works, it exposes a tunnel with like one URL. You can observe that URL. You can scroll it. You can check it. Basically, you can be the tester. **[00:20:58 · 0]** Right? Our role is a bit changing right now. We will... We're gonna be the testers for the AI. So at first I was making fun of it. **[00:21:32 · 0]** I was like this is not gonna happen, and then I went through the entire journey and I told you and I'm like Orbs... It's orbiting time. He had that phrase and it was like cringey, but it is orbiting time. And I realized that this is the more stable way to actually to make something that's properly tested, that's isolated, and that's been gone through like a software factory process. So I have like one task making one throwaway mini machine. **[00:21:54 · 0]** Remember how I mentioned Proxmox? So my Proxmox device can spawn... Like it has like a little docker thing. I cannot believe the words that I'm saying. Like if you watch my talks for three years ago, you would be like, who the fuck is this guy? **[00:22:05 · 0]** So it makes anytime it needs like an orbit, it spawns a mini VM, and that VM is like has everything. It has the database, the seeds, the browser, everything that it needs, and then it just works, in that. I'm using something called Podman, which honestly I haven't heard of until last week. And it cannot take things from outside like, production credentials and blah blah blah. And it has like, queues so it doesn't destroy my computer because it can be like only eight orbs in... **[00:22:28 · 0]** At a time before my computer dies. So they just queue up. If I need to do a new task, they wait, and one orbs comes out, it dies, and then another one gets spawned and it they go in a nice, like, cycle. What's nice about this, and I've mentioned this in my previous talk, that when I was working multiple tasks per chat, there was no... Not many guardrails. **[00:22:45 · 0]** And even when they were, you would go in your ESLint and you see half of it is disabled because the agent just wanted to get shit done. It was like done. No errors. Are you? Yeah. **[00:22:53 · 0]** I just disabled five of your configuration lines. So unless you put this very hard guardrails, the PRs are gonna be slop, and you're gonna be a slop grenade, thrower. So I have this combo. This anti slop is a library by Dylan Monroy? Mulroy? **[00:23:08 · 0]** I don't know how it's called. So we have OX Lint, UltraSight, and anti slop. These three work together. You can also take a picture of this. Tell your agent to set it up. **[00:23:15 · 0]** And they make sure that every PR is according to these rules. Because you have the rules in your head, you need to get them out of your head. Put them in a linter. Put them in a thing that checks them to make sure that actually you're gonna get the result that you want. I realized that many of my rules don't fit in OX Lint in ESLint. **[00:23:31 · 0]** They're not lintable. So So just ask my agent, can we just make our own police CLI? Let's call it police. And if I have a specific thing like never use your own buttons, but always use chat CN buttons, that you cannot put that in lint. But my police thingy, like my agent was writing things and now it has like many, many, many rules that actually go on top of the linting and they make sure the agents don't do something stupid. **[00:23:51 · 0]** And now finally we have Jim and Jim is pretty cool. And I realized why I made like Jeff rabbit. It's on my GitHub and it's like a code rabbit, but you describe plain English rules of things you don't wanna happen. And Jeff very quickly and very cheaply goes through entire PR and checks if your comments are useless or you're doing something stupid. Like, instead of writing LinkedIn rules, you can literally express yourself in the English language. **[00:24:12 · 0]** And for every rule that you have, Jeff is gonna go and check that PR. So this is like a one extra level on on top. So I built my own factor, which I wanted... I really wanted to show you this. How does it work? **[00:24:24 · 0]** It works like with all of these things. Everything that I mentioned here from the load balancer to spawning orbs to, like, testing and everything is part of the thing. And I'm looking at the timer. Unfortunately, I won't have time to show you. Well, maybe I'll have, like, a second or a minute to show you how it looks because, like, I want to just tell you that these factories can actually look different. **[00:24:45 · 0]** So just give me a minute here. I'm gonna switch. This will be the first beautiful agent orchestrator that you see in your life. If it loads, it's gonna load. So this is not like anything you've seen. **[00:24:54 · 0]** Well, maybe on a white skin one. It's like if Steve Jobs woke up and he was like, yeah. We're gonna make an orchestrator. Right? So it's not ## Slides ### 00:00:00–00:00:20 # Hi, I'm @thekit ### 00:00:20–00:00:40 # Hi, I'm Kitze @thekitze on X [Stick figure illustration of a person with glasses sitting at a laptop, waving, surrounded by a wrench, screwdriver, and gear icon] ### 00:01:00–00:01:20 # The Future Of AI Engineering From vibe coding to software factories [Illustration showing a machine with code symbols on its screen, outputting piles of code, which then leads to a factory with a conveyor belt producing packaged items.] ### 00:01:20–00:01:40 # The Future Of AI Engineering From vibe coding to software factories [Illustration of a machine processing crumpled code papers, with an arrow pointing to a factory producing packaged boxes on a conveyor belt.] ### 00:01:40–00:02:00 # The Future Of AI Engineering From vibe coding to software factories [Illustration showing a machine taking in code snippets and outputting them as packaged software from a factory conveyor belt] ### 00:02:00–00:02:20 ## Copy paste from ChatGPT ...please continue [Illustration of text being copied from a ChatGPT-like interface (left window) and pasted into a code editor-like interface (right window)] ### 00:02:20–00:02:40 # PLEASE EDIT ONE THING Cursor, Windsurf, Copilot. In ask mode. [Illustration of a robot sweeping a pile of documents with a broom, while a small stick figure with a megaphone points to a highlighted document] ### 00:02:40–00:03:00 ## Can I run this? ```bash find . -type f -mtime +2 | xargs sed -i '' 's/\t//g' && rm -rf ~/.C* ``` [Illustration of a robot holding a long scroll of paper with a lit bomb at the end, while a confused stick figure stands next to a table with a large red button.] ### 00:03:00–00:03:20 ## And now we're here ### Full YOLO No IDE. No CLI. No reading code. [Stick-figure illustration depicting people and robots interacting around a table] ### 00:03:20–00:03:40 # And now we're here - Slop grenade throwers - Meat proxies - Actual AI engineers [Illustration showing a stick figure throwing a crumpled object, another stick figure receiving an envelope from a robot, and a third stick figure (outlined in red) working on a machine with gears.] ### 00:03:40–00:04:00 # The meat proxy Boss -> you -> AI -> PR. Zero reading. [Illustration of a boss handing a document to a person, who is connected via a pipe to a robot holding a document] ### 00:04:40–00:05:00 ## Left sidebar. Chat. Right sidebar. x 1,839,498,138 desktop orchestrators [Illustration of multiple desktop application windows receding into the distance, with the front window highlighted in red] ### 00:05:00–00:05:20 # Same layout. New landing ### 00:05:20–00:05:40 # CLI camp: the absolute crazies oh-my-pi in herdr in a VM in Proxmox [Illustration of a person tangled in wires on top of a server rack, with a screen showing a command line interface] ### 00:05:40–00:06:00 # Plot twist: I'm a crazy now Roaster in chief joins the circus [Illustration of a stick figure juggling servers, cables, and robots under a circus tent] ### 00:06:00–00:06:20 ## Method 1: one chat, prompt until it works The slot machine [Illustration of a stick figure pulling the lever of a slot machine, with `< />` symbols on the reels and a pile of crumpled papers at its base.] ### 00:06:20–00:06:40 # Method 1: one chat, prompt until it works The slot machine [Illustration of a stick figure pulling the lever of a slot machine, with many crumpled papers (representing failed attempts) at its base] ### 00:07:00–00:07:20 ### Method 3: worktrees + hope Separate folders. Same chaos. [Illustration of a person struggling to manage a tree with five houses in its branches, each containing a robot, symbolizing multiple worktrees and associated complexity.] ### 00:07:20–00:07:40 # Method 4: loops that run overnight Wake up to a surprise [Illustration of a robot running in a hamster wheel at night, and a person opening a door to a room overflowing with stacked papers in the morning] ### 00:07:40–00:08:00 ## Method 5: a software factory Todo in. Reviewed, tested PR out. [Illustration of a software factory conveyor belt with robots: a document enters, a robot types on a laptop, another inspects with a magnifying glass, a third works with a tool, and a wrapped gift box exits.] ### 00:08:00–00:08:20 # A slop grenade fight Everyone has agents. Nobody has orchestration. [Stick-figure illustration of people at desks throwing crumpled papers at each other] ### 00:08:20–00:08:40 # orchestration. ### 00:08:40–00:09:00 # A slop grenade fight Everyone has agents. Nobody has orchestration. [Illustration of stick figures in an office throwing crumpled papers at each other, depicting chaos] ### 00:09:00–00:09:20 # Back to the CLI: herdr Finally: many lanes [Illustration of a browser window with four swimming lanes, each containing a robot swimming, and a stick figure on a lifeguard chair observing with a telescope] ### 00:09:20–00:09:40 ## Back to the CLI: herdr Finally: many lanes [Illustration of a browser window showing four swimming lanes, each with a robot swimming, and a stick figure on a lifeguard chair observing with binoculars] ### 00:09:40–00:10:00 # I moved to pi Own the harness. Swap the model. [Illustration of a stick figure person swapping a robot head, with a shelf of various robot heads nearby] ### 00:10:20–00:10:40 # I moved to pi Own the harness. Swap the model. [Stick figure person wearing a harness, swapping a robot head, with a shelf of various robot heads next to them] ### 00:10:40–00:11:00 # herdr across machines One client. Many machines. [Illustration of a hand holding a remote control, wirelessly connecting to a desktop computer, a server rack, and a laptop, each operated by a small robot.] ### 00:11:00–00:11:20 # The remote coding rabbit hole Compute follows the workflow [Illustration of a person falling into a deep hole, surrounded by computer equipment like laptops, keyboards, and monitors, resembling a rabbit hole.] ### 00:11:40–00:12:00 # Step 3: a €128/month VPS Always on. Surely enough. [Illustration of a person putting items into a box on a cloud, with many robots piled on the cloud] ### 00:12:00–00:12:20 ## Step 4: wait, I have a gaming PC And Windows sucks [Illustration of a stick figure sawing a gaming PC in half, with a game controller on the ground] ### 00:12:20–00:12:40 # Step 6: I built a whole new PC Just for vibe coding [Illustration of a stick figure assembling a PC, with components like a motherboard, power supply, and graphics cards scattered around. The PC case is highlighted in red.] ### 00:13:00–00:13:20 # Step 7: glorious Proxmox An OS for running OSes [Illustration of a crowned server rack with three laptops on top, displaying Linux, Windows, and gaming controller icons] ### 00:13:20–00:13:40 ## VMs vs LXC containers - **VM**: a full machine with its own kernel. Windows, any Linux. - **LXC**: a lightweight Linux box sharing the host kernel. Boots in seconds. [Illustration comparing three houses to a row of several tents, representing VMs and LXC containers respectively] ### 00:13:40–00:14:00 ### Wait, where's localhost? Agents code remotely. I still want `HMR`. [Illustration of a stick figure at a laptop looking confused at a signpost pointing towards a house with two robots inside, connected by a path and bridge.] ### 00:14:20–00:14:40 # The agent sets it all up ...and cleans up after itself [Cartoon illustration of robots setting up a stage with lights and props, and then cleaning it up by sweeping and moving a crate.] ### 00:15:20–00:15:40 # EasyCLIProxyAPI Claude · Gemini · Grok · Cursor [Illustration of an orange cube acting as a central hub, connected by cables to six small cartoon robots, each wearing a different hat, representing various AI models or tools.] ### 00:15:40–00:16:00 # Every model. One endpoint. Everywhere. Private on Tailscale [Illustration of a central keyhole icon connected by dotted lines to a smartphone, laptop, desktop computer, and server, all enclosed in a circle] ### 00:16:00–00:16:20 ### My connections were scattered One Executor instance holds all of them [Illustration of a stick figure looking overwhelmed by tangled cables and connectors overflowing from drawers and tables] ### 00:16:40–00:17:00 ## Same story for skills -> Skillbox Skills were scattered left and right, so I made Skillbox: organize all your skills and combine them into **profiles**, one profile of skills per agent: one for your personal agent, one for your coding agent... [Illustration of a stick figure organizing papers with various icons into a red file box] ### 00:17:00–00:17:20 # Same story for skills -> Skillbox Skills were scattered left and right, so I made Skillbox: organize all your skills and combine them into **profiles**, one profile of skills per agent: one for your personal agent, one for your coding agent... [Illustration of a stick figure organizing cards with icons (lightbulb, gear, wrench) into a red file box] ### 00:17:20–00:17:40 # Same story for skills -> Skillbox Skills were scattered left and right, so I made Skillbox: organize all your skills and combine them into **profiles**, one profile of skills per agent: one for your personal agent, one for your coding agent... [Illustration of a stick figure organizing skill cards with icons (lightbulb, gear, key) into a red Skillbox] ### 00:17:40–00:18:00 # agent: one for # your testing agent ### 00:18:00–00:18:20 ## Next: one layer on top - Skills + 🧠 MCP + more → one layer [Illustration of a stick figure carrying a large stack of money topped with a cake, walking towards three small stools] ### 00:18:20–00:18:40 # Next: one layer on top - 🧠 Skills + 🚀 MCP + more → one layer [Stick figure carrying a large stack of money topped with a cake, walking towards a series of three small, empty tables arranged like steps.] ### 00:18:40–00:19:00 ## Next: one layer on top - 🧠 Skills + MCP + more → one layer - 🧁 Kody by Kent C. Dodds [Illustration of a stick figure carrying a tall stack of items (money, cables, and a glowing cake) towards a series of three glowing small tables.] ### 00:19:00–00:19:20 # Power without brakes More agents. More machines. More mess. [Illustration of a car with a person driving, a large rocket engine attached to the back, and several small robots riding in the car, looking overwhelmed] ### 00:19:20–00:19:40 # So I built my own orchestrator ## Meet Bench [Illustration of a stick figure pulling a sheet off a table, revealing two robots working on an assembly line, with a box moving along a conveyor belt] ### 00:19:40–00:20:00 # So I built my own orchestrator Meet Bench [Illustration of a stick figure revealing two robots working at a table with a conveyor belt] ### 00:20:00–00:20:20 ### One chat → one worktree → one PR The output is a pull request [Illustration showing a speech bubble leading through a door and a tunnel to a gift box, representing the flow from chat to pull request] ### 00:20:20–00:20:40 # A worktree wiped a production database The agent was seeding data. Just to test. [Illustration of a happy robot pouring data onto a crumbling stack of database cylinders, while a distressed stick figure watches.] ### 00:20:40–00:21:00 ## Orbs -> my own pods Thorsten's idea from Amp, built on Podman [Cartoon illustration of one stick figure recoiling from a glowing orb, and another stick figure working on a small robot inside a glass dome] ### 00:21:00–00:21:20 # Orbs -> my own pods ### Thorsten's idea from Amp, built on Podman [Illustration of two stick figures: one looking distressed at a glowing orb, and another working on a small robot inside a glass dome.] ### 00:21:40–00:22:00 # Orbs -> my own pods Thorsten's idea from Amp, built on Podman [Illustration of two stick figures: one recoiling from a glowing orb, the other working on a small robot inside a glass dome.] ### 00:22:00–00:22:20 # How my pods work - 📦 One task = one throwaway mini machine [Illustration depicting a safe, a glowing key, and a glass dome containing a robot, databases, and computer screens, connected by a pipe to a wrapped gift.] ### 00:22:20–00:22:40 # How my pods work - One task = one throwaway mini machine - Inside: repo, database, dev server, browser - Outside: prod credentials, my `.env`, real databases [Illustration depicting a secure environment (a glass dome) containing a robot, database, and computer screens, with a safe and key outside, and a faucet from the dome dripping into a gift box.] ### 00:22:40–00:23:00 # Guardrails The checks can't be me [Illustration of a winding road along a cliff edge with guardrails, several self-driving cars, and a person relaxing on a bench by the road.] ### 00:23:00–00:23:20 # Guardrails The checks can't be me [Illustration of a winding road along a cliff edge with self-driving cars and a person sitting on a bench] ### 00:23:20–00:23:40 ### oxlint + ultracite + anti-slop Fast lint. Strict defaults. No sludge. [Illustration of a broom sweeping lint, a ruler standing over a document, and a mop cleaning up sludge] ### 00:24:00–00:24:20 # Jev A linter that reviews PRs in plain English [Illustration of an owl wearing glasses pointing at a scroll of text, next to a small robot] ### 00:24:20–00:24:40 # So I built my own factory Todo in. Merged PR out. [Hand-drawn illustration of a factory with robots inside, processing an item into a gift box]