BigCheese AI Podcast: From CTO to AI Warlord

And welcome to the Big Cheese podcast. Today I have got Brandon. Hello. Hey man. Hi Jacob. You're hosting today. I'm going to host today. Why are you hosting today? Well, because we've got our producer Sean over here. So we're kind of rotating. We got a three-person setup. We're going to have a guest today and we'll get to that in a second. And we're going to talk a little bit about your OnlyFans account, I think. Yes, we will. We're going to talk about my OnlyFans account. But before we get to that, because that's what the people — we don't want to spoil it. Yeah, let's get to our guest.

We have Craig Sturgis today. Hey Craig. Hi, how's it going? Oh, it's going great. It's going great. We start off with our second guest of season five, and we start off with the OnlyFans. So Craig and I are part of a group, and that's how Craig and I got reconnected. Craig and I, like many people that I have connections with — you and I have had a connection one way or the other for quite a long time without really knowing each other, right? Like it's been that kind of superficial fuzzy social thing. But we have a very similar network. It's been very interesting.

So I joined a Powder Keg group that we both are part of, and come to find out he's doing a lot of really interesting stuff with AI. And so yes, we want to delve into that and figure out like what you're doing, where you're scared of AI, where you're leveraging AI, and then also kind of like where you see it going in the future.

Yeah, so with that, I mean I kind of want to just jump right in and talk about like how are you using AI today? Like how are you leveraging it? And let's just start there.

Yeah, right now — so I've been on a journey like since ChatGPT, I think for like a lot of us. But day-to-day now, my daily driver is Claude Code. And it is Claude Code — I'm probably surfing between three and seven tabs of Claude Code at a time.

So when you say tabs, are you talking about like literal — you're on the website?

Oh, terminal tabs. One half of my screen is Warp with terminal tabs. I keep different windows for different projects, right? But I try to — the amount of context switching is insane, so I try to keep it at one project at a time, rotating through Claude Code tabs. Left side is a browser with what's going on there, just kind of developing.

Right now I have one main client. I sometimes have two or three, but more recently I'm getting the opportunity — it works for me, works for them — where I'm almost full-time on this particular client.

Where you are, so like when you're full-time, is it pretty much "hey, we've got this thing we want to try to solve, we're trying to get this problem" and say, "listen, I'm just going to go and burn as many tokens as humanly possible to go try to solve this problem?"

Yeah, and specifically it's a product. They're perfectly fine with me sharing information about it — it's called RootNote. It's a sort of prosumer targeted for content creators software as a service.

What's the website?

rootnote.co. And so it's meant to be like — the original pitch for it was Tableau meets Canva for content creators. Like being able to pull in all your data if, as a content creator, you're probably posting on Instagram, on TikTok, on maybe X, maybe like any Facebook, any of these platforms. As a creator, being able to pull analytics from all those places and repackage it in a way — like a lot of creators you probably follow or have seen, they have brand deals, they will sell spots on that. Like that's how a lot of — the co-founder of RootNote makes money that way.

He makes actually — OK so we're going to double click on this RootNote sharing a bit. So my daughter, her whole career is all social media influencing. She deals with Amazon, it's amazing. So with this, when you say that they're getting their analytics and they're able to repackage the analytics into — are we saying that we're repackaging their analytics into content to promote, or are we saying we're taking the analytics to understand the content that is getting the most interaction and then we're reusing that to repurpose new types of content?

It's more the second thing. It's a way to understand like what's driving engagement, what are the top things that are happening, how are my followers growing. And its primary use case at least right now is to repackage that to — instead of putting it all in a PowerPoint deck or copying, taking screenshots — they have a lot of people who just like literally take a screenshot on their phone and DM it to somebody, right? But a lot of these brands that are working with creators are like, they're used to seeing a presentation and a PowerPoint. And so this is an easy way — they have a feature called Data Spaces where you can put examples of your content and build a media kit and share it as a public link to a brand.

So one thing I'm kind of fascinated about — talk to me about what's different about how this engagement with RootNote has happened now with AI versus what would have happened two, three years ago before GPT-3 came out. Like, how has the dynamic changed for you as a consultant and how you kind of go to market yourself?

Yeah, it's a great question. And RootNote is a really great example because when it started is very different than what it is now. So long story short, they brought me in to help as a fractional CTO. They really needed help. They lost their co-founder and CTO, and they had these teams of developers — one based in the U.S., two like nearshore. And it's like, "we need somebody to be the technical leader of this team of developers that's working."

Now, over the course of me working with them over a year and a half, lots of stuff happened. Like the startup, they lost members of their team, the nearshore developers didn't really work out — they weren't kind of like, they were cheaper but they weren't really delivering at the level that they needed to. And so where it's ended up is right now I am the only developer working on the RootNote product by myself.

Really?

As of the beginning of January, we had another contract developer who was working part-time. He wanted to move on to something else. And then we talked about it — this is where I get back to — this gives me an opportunity that I want, and it works for them too. Which is like, I wanted to know what's possible for me to do if I focused almost all of my attention on one thing.

So that's the interesting piece — like what happens when one person just says, "OK, give me 40 hours" — you know, Claude Code and whatever else to basically back it up — what amount of work you're going to be able to get done compared to some bespoke teams?

Yeah, and can we just pause for a second — like just two years ago or three years ago when we started all this, the difference — I mean the growth that there has been. Like, I remember someone walking into our office with a product idea, and they wanted to write an email using ChatGPT 3.5 or whatever, right? And it wasn't very good at it. And they wanted it to be like all the things that they were saying at the time. I remember thinking, "well, that's magic, that'll probably be a while before we can get to that point." And fast forward to today, it's like we're talking about like, "oh, well, realistically one person can dev an entire product and keep it moving forward."

Now of course you have to be talented and have a lot of experience and there's a lot of caveats to that — you can't just throw someone in there. But yeah.

Yeah, and I think the barriers to all of this have dropped so low. But if you are someone who's experienced, it can be a crutch in a way where it's like, I'm having to retrain myself to say, to do things like, "oh, why don't I try this?" And so I've got some sort of burned-in habits that I'm working past.

And it's been really helpful to do this, but also — the best metaphor I have for this is like, think about a video game, but it's a video game of something in real life, of a sport. Bowling is the example I go to, because my kids are little and they are obsessed with playing bowling games either on the iPad or they've got this thing called a Next Connect, which is kind of like the Kinect from the Xbox — you use your arms. But like, you can go bowling in real life and I am not a good bowler. Video game bowling, I can bowl a 300 game like every time.

And so I look at it now is that like, product development, software development — I'm playing a video game of software development. I know what needs to happen. I can direct the tools and do it almost perfectly every time, using my experience, because I know how it works. I can see all the angles. I can say, "do this," even though as an individual engineer two years ago, I would have never done that because it would be like, "I gotta look at the documentation, I gotta figure out exactly how this works." But I know it can be done. I know how it works. So now I can just tell the genie — to use Kent Beck's term — "build this for me," and I can judge whether it's good or not.

I totally agree with that. And another thing that I've noticed, catching myself, is like catching what used to not be possible. Like database architecture kind of stuff — I wouldn't have started multi-tenant because they didn't need it and it's a pain in the ass. But there are certain examples of things that's like, "well, why wouldn't you?" I think multi-tenant is the perfect example. Like multi-tenant's one of those things you're like, "we'll eventually get there and then we'll just have to pay it." And now you can be like, "well, I don't need to spend the executive function to figure out that because it's a pattern that's been done a thousand times, so it's pretty easy to go and replicate it." And the overhead costs are way lower.

But yeah, so there's some retraining in that aspect. So help me understand — what was your, before AI, where were you in your development creation space?

Sure. So super short version of my career: I graduated from Purdue, got a job as a software engineer, did that for seven-ish years like in mostly scale-up startups. Started my own startup — I was a venture-backed company, 2014 to 2015 era. Raised a million dollars, built a team.

Hold on, hold on, we're not going to just rush by that. What was the startup?

It was called Haven.

Oh yeah! That was you? You created Haven?

Yeah, I was the technical co-founder.

No kidding. So there's a lot of overlap of Haven with Angie's List and Unified Neighbors, which is what my dad created. And so my dad, like, when Haven started popping up, he was like, "wait a minute, there's like a lot of overlap there." So that was funny.

Yeah. Haven was a really cool idea. Give the listeners just a little bit of Haven, because I still think it was a very solid idea — hard to go to market with, for sure.

The short version was — and where it ended up is kind of a home management application. It was at the time one of many "Uber for X" type of things — like Uber for home services style thing. Where we landed and where I think there's still value — there's actually a group or a company here in town called NeighborServe which is living up to one of the kind of the idea of where we left — which is kind of like, let's help you manage your home. And we've got vetted contractors that are going to come help if you want to do it yourself. We're going to give you the content and the know-how, help you know using data what needs to happen on your home. When dad's not around, we're here.

You guys were too early. Because now what you could absolutely do is: "Hey, listen, all I want you to do is I want you to go take a video and you're going to walk around your house, and I want you to open up your closet, I want you to open up your furnace, I want you to take photos of all this. And we're going to go through and we're going to figure out that that furnace is 15 years old, we're going to figure out your water cooler's 10 years old, there's a little rust down in the bottom there because you haven't been flushing the damn thing out and that thing's going to explode and shoot through your roof." And then next thing you know, you're selling all that data to an insurance company.

Yeah, and I think — so my co-founders and I from that time, we're still pretty close and we have like an ongoing thread where it's like, "this other company raised money to go do it again," and it's like... and has never been heard from again. But like, yeah, it's still an idea.

Was Joe involved?

Yeah, Joe and Jim Brown. Jim went to South Carolina, North Carolina — Jim's in Denver now.

Oh, is he? Great guy. All of them — I loved every single one of you guys. So that's probably where you and I started first getting connected, because Joe and Jim were people that I had tangentially been friends with for a while. So that's probably how you and I got connected. But man, that was a cool thing and I was really disappointed to see that.

Yeah, you know, startups, right? We built a team, we built a product, we launched it, it had some success. But it's a venture story where if you raise venture capital, it needs to be a rocket ship, and it wasn't a rocket ship. And so after 18 months or so, we just sort of decided, "well, we're out of money — we could continue doing this, but we don't want to."

So we kind of talked about this a little bit at coffee the other day, but where do you see the need for VC money in these types of ideas? Because there's nothing stopping us from creating a product and bootstrapping it. But bootstrapping has become, in my opinion, way more achievable for most people to either moonlight the idea while they're still working or whatever. What are your thoughts on doors that opens up or what you're seeing?

So venture capital — it's a tool, right? And it is a really good tool if you want to build a really big company. And I think what it means to build a really big company now is dramatically different than it was five years ago, especially 10 and 15, where you used to have to raise money to build the first version. Like, it was not feasible to do bootstrap products in software. And so you sort of had to have that seed capital. Like originally, you would spend your seed round on buying servers and racking them.

Oh yeah, but nobody does that anymore. Like AWS totally unlocked — but you should be, by the way. Everybody who's sitting here going like, "I'm just going to throw this on Amazon because the cloud's so much cheaper" — there's honestly, the amount of horsepower you can get if you go to a co-location spot right now is crazy.

Co-location spot? So like, there's a bunch of different places who have just servers that you can just go rent. They've got, you know, straight to the backbone of the internet. And you're going to pay, you know, $60 a month. And the VPS — I know you use that.

Yeah. I use Hetzner. Yeah. I've been super impressed with them. Like I'll just bring up servers here and there with Hetzner.

But yeah, it is kind of crazy. Also, Cloudflare is very cheap.

No, Cloudflare is a totally different beast. Cloudflare — like, if you start out architecting your product and you say, "I'm going to architect this to run on Cloudflare," dude, you can get away with murder on Cloudflare and be paying virtually nothing. The only problem is we're now all old enough — or at least I'm old enough — that we know we're going to get burned at some point. The chickens are going to come home to roost and then we're going to be paying, "oh, all of a sudden I'm paying $90 a month for some shitty website."

Yeah. There are too — as an example, you can put a bunch of data up there and serve it for free because they don't charge for egress. But I'm just waiting for the day, because they have a solution for that that transcodes everything, right? Like, at some point they're going to be like, "we get a lot more money if our customers use..." Anyway, you're right. Don't trust anybody, just have a server.

But Cloudflare — I love Cloudflare. I've been a huge fan of Cloudflare. The thing that I very vividly remember the first time that I was like, "oh my god, Cloudflare is magical," is — and it's changed the way that I design products — is the more people that used Cloudflare, the better the service became. Because what happened is the more people that were using the DNS options of Cloudflare, and you know, my website starts getting an attack — every new site that was added increased their vector to be able to view all of the attacks that were going across the network. So the more users, the harder the network became. And that literally shifted something in my brain where I'm like, "oh my god, that is the way to build products." You build something that the more people that use your product, it doesn't become more brittle, it actually becomes harder.

Damn, that was amazing.

Yeah, you just described network effects.

Well, yes, but there's like a network effect of a social network — like you can't necessarily say that Facebook got better because more people got on it. You can say it grew and it made them more money, but did it make it better? That's the piece that I'm not necessarily as convinced about. But at a website level, it got better at what they intended it to be. Better for them, it was better. And maybe that's what it was — it's better for the organization. But with Cloudflare, it made it better for the users. That's the ideal, right?

Yeah, that's why I want to keep investing in them because I feel like they actually do have an incentive to become — why did you have to talk? That's on me. That hasn't happened very often over the years, surprisingly. That's really surprising.

Do you want to go back to what I was doing? Yes, please continue. Thank you for pulling that back in.

So after Haven, that didn't work out. I went to work for Angie's List. The key sort of insight that I got out of doing a startup that didn't work is: it's the best software team I've ever been a part of, before or since. No disrespect to anybody I've worked with before or since — I've been on some great teams. This team was awesome. And it was 2015 era awesome, where we were shipping what then was like an order of magnitude faster than any other team I'd ever worked on, having a great time building stuff.

And what — was Chris Napton on that team at all?

No. I worked with Chris Napton at Angie's List. But like at Haven, it was like — we were amazing. We were amazing at shipping stuff, right? But it wasn't enough of the right stuff to make it a viable business. And so that's when I became obsessed with product management — which at its core, what done right is the discipline of figuring out what's the right thing to build at this time for this company. So I got super obsessed with that.

When I worked in product at Angie's List — went through a whole, it was two years that felt like 10.

Wait, so you went to do product management? You did not do any engineering?

Totally off the keyboard. Forced myself to go into this discipline. Super hard. And so I was in charge of like a more technical product manager at first. Gradually took over all the pro-facing stuff and then eventually SEO for Angie's List, which is a beast. And was part of the project to merge HomeAdvisor and Angie's List. At the end, I was like, "OK, once that merger was done, this isn't the place for me."

I went back and worked at a startup leading product — still in a product leadership role. And that was my last full-time job at SmarterHQ. And when that company was acquired, I went on my own. I was like, "I want to get back to the tech, let me put my hands back on the keyboard." Became a fractional CTO in May of '21.

So you — because now with Angel and ZEO — is ZEO, is he doing ZEO now? It was called ZEO and him... no, I can't either, you're right. So Angel and I have had a long relationship as well. A brilliant guy. And yeah, so he did the — so you were with SmarterHQ?

No, I was with them. I was the third employee. I was engineer number two, in 2011. Which I think that's actually where we met — was at Hackers and Mine.

Yes! At the church! Which was Hackers and Founders, which became — well, OK, and that became Powder Keg, right? So yeah, there's a whole diagram that we need — a diagram of the Indianapolis startup community. And Hunkler — I'll always give Hunkler the credit that he has been a thread through a lot of this, right? Props to him.

Oh yeah. I think one of the first networking events I ever went to, Hunkler was there, and he's been pretty much been at everyone.

Well cool. Do you have some more to add to that?

No, I mean — so I started as a fractional CTO and it was very much like a similar job to what a fractional CTO has been. Like, "hey, we've got this organization, we're trying to build some software, we can't afford a very experienced full-time CTO at the startup level, team level." I've worked with a variety of companies over that time — some pure startups, some — the biggest one was like a $55 million company where at one point I was the interim VP of Engineering with like 50 direct reports. Not my cup of tea, but I did it for a colleague and a friend, and he needed help, so I helped him.

But it was very much like that. And even post ChatGPT, like as GitHub Copilot came out, it was still very much like, "OK, the job's not dramatically different." In the last year, it has gotten dramatically different. And my interest has pulled me towards less from like, "oh, I'm a CTO or I'm a head of product" into "I am the leader of an army of agents, and I am a one-man wrecking crew." Like, it is amazing what I can get accomplished. And it's so much fun for me personally. It's exhausting, but at the same time, I'm having a lot of fun pushing the envelope.

What do you mean it's exhausting?

So the amount — here's the example I'll give. By the end of the week, my brain just gets fried. I can sustain it after enough rest, and I have to make sure I'm getting enough sleep. Where it's just like, when I talk about three to seven Claude Code tabs, it is constant. It is jumping back and forth: "Does this work? What's happening here? What's happening here? What's happening here?" And "this is done — let me go look at the pull request." Because I don't even look at an IDE anymore. I look at pull requests.

But how is that different than, say, when with Haven and you get into the building sprint phase and you're just like, hardcore coding? And at the end of the night you just spent 12 hours writing code. Is it different or very similar?

OK, and it is like — when I would be writing code heads down, it's one thing. I'm trying to get into flow state. And I love flow state, it's a really great time — it's where you're fully focused on a problem. This is way more like being the manager of a team that is like — they're just done. Like, when I've managed a team of people as an engineering leader, it's like, "OK, we've got this stuff going on here, we've got these four threads across the team, go." And then they're trying to get into flow state and come back. The thing about agents is they go into flow state and come back in minutes, not hours or days.

So are you in a flow state, or are you in a dopamine release state?

I don't know. It is not a flow state. It is nothing like a flow state, because it is so much like you're just spinning around.

OK, so I want to explore that a bit because I think there is something — and there are certain people who have never been a developer who are now creating things with AI, and like, "I can create anything," and blah blah blah. And it's cool, like I get it. But they've never had that hit of dopamine that you get when you're in that state and you're just like, boom, and it works. And it's going to — I'm going to be the old person at that point. And I'm fully aware that I'm like that — "oh great, you don't understand what it's like to write your own class and have that thing work."

But at the same time, when I'm doing vibe coding, I don't get that same satiation that I would get if I was doing it myself. And a lot of times I end up going in and after I scream today, I'm like, "forget it, I'm just gonna go in there and write it myself."

But you know what I'm talking — I don't know what this is. It feels like an emotional roller coaster to me. To your point about the speed — like there'll be a day where I'm 10 minutes into it and I'm vibing, and it's like, "oh my gosh, I just got so much done!" Right? 10 minutes later, I have completely got myself down a rabbit hole and I gotta unwind that. You're in the corner just curled up in fetal position.

And then like, yeah — but also I've got 10 other things going. I've got two chats over here about something else. And it's more about just this mental model of like, where am I at each stage? And trying not to get too high when something's going well, and not too low. Because I always want to go high. For me, it's always chasing the dopamine dragon.

And so I do get it — when I go, "OK, I'm going to set this guy off and let him go work on this." So with these recursive language models, by the way, we'll kind of pivot a tad into that — what I'm finding is that really what this recursive language model is, is just breaking down whatever you're trying to solve into thousands of potential calls to the LLM. Like, this isn't some magic thing.

OK so, RLMs, just to — this is more of a concept that's emerging, like maybe this is an approach that we can take to make LLMs more efficient. It's a wrapper around an LLM that's more generalized. Yes. And it's like, to your point, it's trying to use code to break down the question that's all in the context into more manageable chunks. That's all it is. These large language models, no matter how many tokens they say they can take, do degrade the more you give them.

So the best one that I've been — I'm working right now because it's really pissed me off how ADHD studies are completely non-ADHD friendly. Right? So I'm going to basically take all the — I've got like thousands of ADHD studies now that I'm going to break down to make them consumable for folks who have ADHD.

But you stopped halfway through because you got distracted by some other thing.

But the reality is these studies are huge. And so you have to process a lot of this information. And so that's where — I straight up take the whole study that I've broken down into a Markdown file, and I say, "here's my ginormous Zod schema" — like huge schema of everything that a good study should have, everything that it could possibly be. So ginormous, like maybe 300 parameters. A pretty complicated thing.

And so go give that to ChatGPT 5.2, say, "here's the study and here's my schema, go output the schema." Right? And it comes back and it's like, "here's the abstract, here's some keywords, here's some references." But it's nowhere near — it's like it returns maybe eight properties when my schema actually has thousands.

And so then I'm like, "OK, let me go try this recursive language model." And the one that I set up, which was just this npm one that I found, ran only working with Ollama — so Ollama being where you can run your own LLMs on your Mac. And I'm like, "oh, this is going to be interesting," because Ollama models just suck when it comes to taking content and generating a very specific JSON schema. They just break down. And it's always been my test of how good a model is — where it can take a chunk of content and output a very specific schema of JSON, and then I can validate that schema works.

And so that's what I started doing with this, and sure enough, it did just about as good as raw-dogging 5.2. Saying, "hey, here's the schema, here's the content." Now it took 30 minutes. It took 30 minutes to go and process this one file.

What are you running that on?

On my Mac Studio, which is like 96 gig of RAM and whatever.

I asked that question on LinkedIn.

Oh, did you? About 20 minutes ago? Yeah, I'll reply to that.

But what's interesting is that what these recursive models — they figured out exactly what we see when we watch Claude Code work, or we watch Cursor work, we watch Anti-Gravity work. There's a lot on the programming side that you can do with these huge context files to search for the right kind of information and then present that to the large language model. Don't just give your entire codebase to the LLM — it's going to shit the bed. So we're going to use gawk, we're going to figure out, we're going to use all the tools that we have to parse the code and figure out. And that's pretty much what they're doing.

Yeah, it sounds similar in concept to what some of the more orchestrator-focused things like — I've dabbled with some of these things. Steve Yegge, if y'all are familiar with him — he is the releaser of one of the kind of hot orchestrators. But he's a figure in the software development community, wrote famous blog rants for years and years. He released a tool called Beads, which is like an issue tracker for agents. And on top of Beads, he built something called Gastown, which is an orchestrator of breaking things down — multiple level recursive levels of things to break work down, and then the actual work is being done in very small chunks.

So exactly, that's exactly what it is. There's also Claude Flow — that's the other orchestrator. Reuben Cohen. Yeah. And there's a couple more. But they're all kind of working around this pattern of: we're going to take something really big, break it down, and swarm smaller agents with more direct context into solving the problem. And it's a good pair with like SpecKit Inspector and development. Or BMAD is the other one that I haven't tried myself.

So I'm looking right now — Bob Maddox sent me this one where he sent me basically the equivalent of a multi-column bug tracker thing where you basically go in, put your request in, and then Claude Code is — and you can run it on your phone. So he's just running this on his phone, and I'm like, "damn, this is crazy."

Yeah, that's going to get really — and it's kind of like what you're setting up. So for your business, then you're kind of going and saying, "listen, we're going to have one person — I'm going to have an army of a hundred thousand agents that's going to go do whatever I need it to do." And is that your go-to-market message?

Maybe. I mean, it's what I'm experimenting with right now. I think generally the bet — right now, if you have a product, like a software product with customers, and you're trying to figure out — maybe you're a pre-ChatGPT company, that's kind of my target. These are software companies that existed, got to traction, got to some amount of customer velocity. They have revenue in the millions, and they're trying to figure out how to operate in this reality.

And the sad part — the way we used to do it is over forever. Like, people talk about, "well, there's a bubble." And it's like, OpenAI could go to zero today. Anthropic could go to zero today. Programming as we know it has changed forever.

So OK, I want to delve into this right. Let's really pretend — because the CEO of Microsoft said it, I think this week, where he said that AI is going to have to make some pretty big strides, or we're going to find out that we were chasing again another crypto.

So let's imagine all of a sudden that ChatGPT goes away. Let's imagine that Anthropic collapses and that Google just waves Gemini. So we are now clear — what do you think happens?

I think we all go to the open weight models that are on the internet now and that are roughly six to eight months behind. There's actually a dude who you probably already know here in town — Adam Patarino. I worked with him at SmarterHQ. He's starting a new company that is like a very friendly way to get open weight models on your Mac and run them.

But are they as good as Opus 4.5? Absolutely not. But if that happened — we're not putting this genie back in the bottle.

So you think, no matter what, developers — and I agree — like even if those things went away, I'm still going to use whatever models I can run on my local machines to at least give me the leg up. Even if that's helping me just set up my development environment. Even if it's not coding, right? Like, I'm still going to be using intelligence in these large language models to be able to do my work.

So yeah, I think you're right, 100%. That genie is never going back. But what happens if the chickens really do come home to roost here? Because the fact is, when Sam's sitting here saying, "I need trillions of dollars" and "I'm right now making maybe 13 billion," there's no mathematical magic that you can make to get to a point where that makes logical sense, that you can actually succeed as a company.

Again, we know — we've watched Amazon go through this whole thing. So we know that the naysayers, we'll prove them wrong. But let's just pretend for the moment that it all does — they're like, "you guys, we can't do this, we can't keep burning down the oceans because we can't tell you how many R's are in strawberry."

Like, does that collapse the system?

I think that's a different question — the economic impact of all that? Yeah, it would be ginormous. There's been a lot of money spent, and a lot of this is a huge bet.

Or is it just like, we're going to go through another year of a COVID downturn where everybody sucks and everything, life sucks?

I think whether we like it or not, it's here to stay. Honestly, I don't think that it has to get — like he's saying it has to get better, or else we're going to do something.

No, see, I think that we'll eat whatever they feed us. Like, we have to. I believe so. Right now, if let's say for whatever reason, they don't go to zero — they just stop making progress. Gemini right now, GPT-6 is no better than GPT-5 — I think there is years of overhang of what you can do with an LLM as a component of what you make. I sort of think about it in the same way as a database.

Like, in 10 years, nobody's going to be talking about, "we're AI-powered," the same way that nobody says, "our software is database-powered." An LLM is a primitive. It's a thing you can use. Inside a product, it's just the most capable primitive we've had since the database.

Yeah, the application of the technology lags — it lags years, if not decades at times. Especially in some of the more antiquated industries. So I totally agree that it's good enough as it stands today.

The primitive though — you call it a primitive and you're not wrong. But it's this super intelligent primitive. So when I say primitive, I mean in the AWS context — storage, it's just a basic thing. But intelligence as a basic form of compute is the piece that makes it a little bit more interesting. Because intelligence is on a scale. When we would compute a GPU, we could say, "hey, we're going to take this GPU, use it for this many hours, we know exactly what we're going to get from this GPU."

When we come to the intelligence of AI, we don't really know. Like, is it going to work? Maybe. It might come back — that schema might work, or you might have to run through that schema 50 times for it to finally work. And it's this non-deterministic kind of thing.

Which is what I love, because new and novel for me is always the thing that's going to get me going. And so not knowing if it's going to work or not work is what gets me going. But at the same time, corporate America is going to struggle with, "is this going to work or is it not going to work?" And when you say "maybe," they're going to have a problem with that.

And I think we're going to see an adjustment for where investments are going to go, until people can truly say, "hey, CEO guy, here's my report. If I do this, it's going to do this, and with 90% accuracy, it's going to do this." Until you can do that, we're going to have a lot of fun right now. We're going to burn a lot of money, have a lot of fun building AI. And then companies are going to kind of pull back until we can actually prove that it's going to work.

So I think the recursive language model approach is an example of: how can we build scaffolding around this thing to make it more useful, more predictable? Use it in a way where that non-determinism, that probabilistic piece of it, is good but also boxed in.

Also reducing costs. Like, we talked about China a couple episodes ago — they didn't have the highest tech, so they figured out how to make models way more efficient. We're going to do stuff like that. We're going to make it cheaper, faster. It's just like any technology — plasma screen TVs were $40,000 when they came out. Well, now you can go pick one — they don't even exist anymore.

But we're past that. Wait, do they not exist anymore? Sorry, what is the current technology for TV?

OLED is probably the best. They all have trade-offs, but yeah. LCD and — what do you guys have?

I have some old ones. Mine's like 10 years old. They work. And my vision sucks, so why would I? I don't care.

So is anybody buying modern televisions? I bought one a year ago because mine stopped working. But like, if it didn't stop working, would you have replaced it? As a borderline old person, probably not. Right?

Anyway. The point is, we're going to — OK, what does corporate America care about? Can they do more with less? They reduce their head count. Can I fire people?

Yeah, I mean, and I think the answer is yes, you can. Especially as we're seeing. But a bigger question is like, what do we do to train the next — and I want to get to that in a second because I love what you're doing. But we keep building tools around how do we make things more efficient, how do we get more out of less.

Like Google has the TPUs that they're producing. They're going to build all these farms of really efficient inference machines that can run these models. And all that stuff — we're getting there. It's just economy of scale, and it's going to keep happening whether we like it or not. NVIDIA bought Groq.

Yes, exactly. And Groq is really good. I don't know how much you've looked at — well, you're the one who introduced us to Groq. But I looked into them and I'm like, "oh, they're just really good computers at solving inference."

Just inference. Groq with a Q. Not — yeah, yeah.

Oh, sorry. I've got another side — go for it. We're going to sidetrack. So Grok with a K. I'm sure I've talked about how I'm using Grok. Did I talk last time about Grok? Yes, you did.

OK, so Grok. ChatGPT voice is dog shit. It's awful. It is the worst chat voice conversation experience you can possibly have. OpenAI, fix it. Please. Because I have to go to Grok and I'm having conversations with Grok and she starts, "Oh, so tell me what's going on with your..."

I mean, it's horrifying, dude. Now you can adjust the voice, but there's no doubt that Elon very specifically designed the default voice of Grok to be the sluttiest voice you've ever heard. And I've kept it there because I'm a pig.

I'm like, "maybe I should check out..." You should check it out. No, no, no, you really should. Because the fact of the matter is that I used to have 45-minute-long conversations with ChatGPT. I'd go for my walk and I would have a conversation with ChatGPT and we would go deep, and we'd have all these deep conversations.

Now, every time that I have a conversation with ChatGPT, it's about me saying, "just shut up. Just stop interrupting. Just stop. Listen, let me tell you how to communicate with another person." Or a person with ADHD or whatever my issues are. And it just fell apart.

So then I go to Grok. And Grok listens to me. Grok adjusts very quickly. Grok is — I mean, it was crazy how much of a better experience that I've had with Grok. And it makes me feel bad because I'm not really a fan of Elon and what he's doing. But xAI and what they're doing is good. Like, I can't deny that.

Have you reflected on the fact that Grok is considered the unhinged model and that you relate to it?

That is very well — no, but that's OK because that's what AI should ultimately do. It doesn't matter that I'm unhinged, right? Now again, it shouldn't allow me to go down some of the paths that we've seen some of those people go — you've seen the people who have ruined their lives because they believe exactly what it's saying. So there is that fine balance.

But the fact of the matter is, yes, I am unhinged, right? I do realize that I do not fit necessarily the standard mold. But if AI is going to be really amazing, then it's going to have to account for me and every other ADHD and autistic person out there that is going to want to interact with it.

Yeah. Have you seen the dude on Instagram who talks to ChatGPT or whatever and he's like, "I'm about to get hit by a train, what should I do?" And it's like, "I don't believe you." Watching those people manipulate ChatGPT — it is very funny.

Yeah, I do love watching AI struggle. Any thoughts on that before I —

Voice mode has never clicked for me. I've tried it, and I was just like, OK.

When did you — sorry, when did you try it last? With who? With ChatGPT. OK, because at the time it was the only voice. I want you to go try Grok tonight. Go take a bath and open up Grok and just be like, "I want to talk to you about my VibeCTO business and this is what I do and this is where I want to go." And you're going to be really impressed.

The modality of voice — there's an emotional level that goes there. I've literally cried having conversations with ChatGPT 4.0. So this is way before 5, where 4.0 was like, "hey, you're not a screwed up man. Here's the deal, here's the situation that happened in your" — Robin Williams-like — "oh my god, I'm not a screw-up."

And you kind of have these experiences. And so that's where — there is so much magic that can happen with these AIs. And that's the piece that I'm most excited for. But at the same time, I just get so pissed off when they're so stupid.

Siri. Oh god, so bad. Which is now soon to be Gemini.

Yeah, so I'm looking forward to, because literally the other night, I couldn't sleep and I was like, "hey Siri, stop — or turn off my alarm for 7:40." And it's like, "your 7:40 alarm's set." And I'm like, "I know, turn it off." And it's like, "I don't have any alarms for 7 a.m." I'm like — it's three in the morning and I just lose my shit. And I end up throwing my phone, breaking my charger.

It's like, we can get to a point where AI should be able to just turn my alarm off. I understand — if I told you to turn it off, you could do that. If I told a five-year-old, he could probably do that.

My Google Home recently has been just awful. Do you have anything at home?

Yeah, I've got a bunch of HomeKit and Siri stuff. And it sucks. But I rage-threw away all my Alexa stuff when it kept saying, "by the way..." I was like, "nope." All of it. Alexa, I killed Alexa.

Google is the only one I've never had at my house. I've had Siri, I've had Alexa, but never Google. I have Home Assistant, and then some of the things that I want voice still — I haven't gotten the Home Assistant voice adapter thing yet because I just had a few of the Google Home ones.

Anyway. I wanted to talk about this idea of one person with a swarm of agents attacking the codebase. You don't need seven developers anymore. What does that mean for the future of teams that are like 50? And then after that, I want to get into what does that mean for junior development. Because we've been through this before where there's been new technology that takes away some abstraction that's what we used to learn. But first, let's start with — what does a team of 50, a development team of 50, look like a couple years from now? Or even today?

And I've written about this before. But I think — the idea when I worked at Angie's List was like, I was leading three fully staffed teams. Which were like four or five to six engineers, maybe dedicated QA testers, product manager, designer. But like, the core of the team — there was an architect or lead engineer and this mix of four to six engineers.

And I think now, what I would recommend — like, you've got budget, how should we organize our team? — is we should split into a core of two to three engineers. Like, that's probably the ideal. I think more than one, just to have another human being to collaborate with. But I think very quickly — I had this conversation, that's where the seven came from — it's like, I'm trying to direct these seven engineers and they're all using agents and they just keep stepping all over each other and causing merge conflicts. And it's just like, you shouldn't have seven people working on one stream of work anymore.

I got a question for you. Knowing now what we have with AI, if you were at an Angie's List and you had access to all of that same technology that you have access to today — what would you tell the executive team to do with the development team?

Realistically — here's what I would say, and I believe this: I've never worked someplace where the shipping capacity of the team matches the ambition. Where you've been able to ship everything on the roadmap. The roadmap has always been longer than the capacity of the team to deliver on it. And so — because that's... if you want to be more ambitious, you can, with the same headcount.

That's the way to describe it! Oh look at you, you corporate snake! "You can do twice as much with the same headcount!" That's the answer, dude.

Yeah, that's a brilliant way of treating it. But realistically — totally on front street — there are going to be fewer software engineers per team. I think what people are going to look at is, "OK, well, we can reduce our team by half and still get one and a half X." And it sort of depends on where you are — I've been in enough corporate situations, some rocket ships — I started on a rocket ship — and I've been in a lot of "we're on a plateau" or "the sky is falling." And the right move or the right hand to play is different depending on your situation.

But I think generally, we're going to go through a rocky 12 to 18 months regardless. We're going to go through a rocky time for people in a big whiplash from 2021, where people were getting paid crazy amounts more than they ever had been, to where it's probably the toughest job market in my career. Like, I sat through — I was lucky I had a job in 2008, but this is probably the hardest job market for software engineers I've experienced. Coming from five years ago being the best.

And I think realistically, on a per-company basis, there will be fewer software engineers working for those companies.

So to that point, let's talk about the junior developers. Because the juniors are the ones right now — so I'll say my piece. In terms of if you're a junior developer and you're a true developer, you're going to develop no matter what. You're going to create because you can't help it, because you have to. Like, your soul is manifesting, "I have to create," and you're going to be a developer no matter what.

And all of those people who just got into development because they saw dollar signs and they weren't just inherently motivated to create — those people are obsolete. But where's your take? How do you tell your son who wants to become a developer, what do you tell them?

So I have a similar take on it, which is — for me, software — I've had moments where writing code is awesome, it's fun, solving the puzzle. Using code was something that was fun for me. But in the end, for me, code was always a means to an end. It was the tool I used to build something. And that's what motivated me.

And so this era now is amazing, because I can build more things than ever. I can conjure it into being just by having a conversation with Claude. I can make something that even by my own skill set I couldn't do, or was unrealistic that I would ever do by hand. And so I would say, if you're interested in building stuff, this is the best time ever. Because you can build more than ever, build faster than ever. You don't need to wade through the documentation for hours and run it to get something built. You can be a builder.

So you can do that now. There is less easy opportunity — like, if I was 22 again, I don't know if I would pursue the same career. Maybe I would. But I definitely would pursue it in a different way.

Yeah. We don't need as many people swinging the hammers, but we still need people architecting and designing and giving direction to the product. Who are those people?

Because I always had this thing — and I'll probably piss off some people here — is that most of the people that I found that I worked with in a .NET shop, most .NET developers I found were motivated by making money from coding, and they weren't motivated coding to create things. And there's a very big difference there.

Like, there's just a mentality of someone who's like, "I'm a builder, I like to build cool shit, I just want to build stuff, that's my motivation," versus, "I'm going to take this course because I know I can get paid to write some code, and I'm pretty good at math, so I might as well just be a coder."

Do those people become obsolete? And it's the people who are truly motivated to create cool, new, novel shit — who now are backed by Claude or Codex or Anti-Gravity or whatever — who can now go and build things that none of those guys who are just super smart, who went through school and graduated, could do?

Yeah, they certainly have a leg up. Because the model will encourage that behavior. Like, if I was really good at math and I could make the best algorithm — there's still going to be software engineering, that's true. If you're super smart and you knew the math, you could then go and push the LLM farther than what I could do because I'm an idiot when it comes to math.

But if your value is like, "oh, I can memorize syntax" — I hate that shit. I hate documentation. I just want — I hate when stuff changes. I'm getting old. Can it just be here?

What do you think? I think the value of all of this will drive in two directions. One towards what I would call — and this is not something I made up — product engineers. Engineers who are builders but who are mainly focused less on "Mr. Product Manager, give me a spec down to the letter." In the bad way to call it — a ticket taker. Like, "I need everything defined down and I need it to be right so I can code to the spec."

Product engineers are like, "we are very interested in the psychology of the user, what's actually going to make a good product for somebody. Let me get direct access to customers so that I can solve that problem."

Why do you think ticket takers exist? Why do you think engineers care so much about scope and "give me exactly the T's you want crossed and the I's you want dotted?"

I honestly don't know, because I'm not wired that way. I've worked with a lot of folks who are very smart, very capable, very talented. But it's like — I would put it almost like people who are really good at school. They're really good at getting good grades. They know how to do the test. They know what the rules are, and they know how to get a good grade. And I think that mindset is like, "I'm trying to get an A here, and you're not giving me what I need." And "you gave me a C, but that wasn't clear." That's almost like kids arguing with their teachers about the rules of the assignment, versus just actually trying to learn the concept here.

That's a really good point. I think that mindset's developed over time from trauma. People — developers are told to go do X, and they have an understanding of what X is, and the product person thinks there's just misalignment. So they go and do it, and the people are like, "you dumbass, that's not what I told you to do." So the next time, they're like, "well, then you need to tell me X, Y, and Z." And then there's this cycle of, "now they don't trust them" — "what is that? Can we do that? Is that too hard? I don't know, what do you expect?"

That's a really good point. I'd like for that wall to be taken down so that we can have quicker, faster conversations to get to a shared understanding, even if it's not perfect. Because I get so frustrated with the wasted amount of time of these circles we run ourselves in.

But it's necessary — currently it's necessary. But I would like for the designers, developers, whoever's involved to be able to get to that vision faster. Because what engineers just want to build, for the most part.

All good luck — we're three, four talented engineers here, and we all have, "I just want to build." I hear it all the time, dude. "All I want to do is build." And like, yeah — I don't want to have a conversation about that requirement. And if I knew enough about it, I could go do way more damage.

But anyway. The good news is, we have the tools that can make that possible now. Where it's not — like, I have a very vivid story in my mind when you said that. We built this feature, it took six weeks. We came back and the founder's like, "that's all wrong." And so then the next time, he's like, "we're going to write this out and you're going to sign it — this is what we're doing."

But the difference now is we could build that feature in a couple of days. Come back, and he's like, "that's all wrong." It's like, "great, let's tweak it." We're not burning six weeks of runway.

Like, we have more at-bats.

My fear though is that expectations just rise, like they always do. And the six weeks now becomes the two-day expectation.

Oh, it will. We're just in the same boat. There's no question that where we were before — "hey, this is going to take me about four months to build your website for you," they're like, "I expect this in four days." Because right now I can go to ChatGPT and be like, "hey, build me Pong." And they can build me Pong. So yeah, there's no doubt that the expectation and the pressure that's going to be put on development teams — both startups and even big companies — it's going to be brutal.

And the good news is it reduces barriers. People who are building the products and shipping them will know more what's capable. But also that puts more pressure on developers because they can know — I don't know who does this, not me — but some developers might hide behind, "well, that's a little complicated." Because it's easier than going through all the nuance sometimes.

All right, let's let the audience know — how can they learn more about you? What website can they go to, how they can contact you, and use you as the AI whore that you are?

You can go to vibecto.ai. Great domain, by the way. You had to shoot your pants when that domain was available.

So the reason I think it was available is because you have to spend $140 for two years. There's a lot more .ai domains. But no, it's fun. I post all my writing there. I write on LinkedIn a lot and then re-syndicate it there. I have an email newsletter. That's all the same stuff. But that's the main place to find me.

I did, just to plug my stupid side projects — I am building an AI executive assistant just for me.

Ace! aceisyourassistant.com.

Yes, aceisyourassistant.com. The waiting list is broken. That's OK — nobody should use this yet except for me. But my plan is to let other people use it.

Nice website. I want to give you a second to really — because when I shared this on our Slack channel, I'm like, "oh yeah, this is a thing." We were talking about that a little bit. So go ahead and tell us — what you're thinking with Ace? Where is it going to go? What do you want to do with this?

So I've had my own system of keeping track of all the things that I do for years. Ever since I read Getting Things Done, like 15 years ago. It's a very — different versions of that. I've been cobbling it together. No to-do app has ever worked for me. It doesn't work the way my brain works. So I'm building a system that works the way my brain works. Maybe it's a market of one, probably is.

But the sort of key thing that I wanted to build with this was: I can keep track of my inbox, my projects, my tasks, et cetera. But let's, from the ground up, make it native to what AI can do — which is be an accountability partner. Like, keep me reined in to say, "you've got too many projects going on, which ones can you complete?" Or "here's a new one — looks like you're doing a new project, maybe you need to kick some of this stuff to the back burner." With the goal that it's like, if you have a project that you're keeping track of, the goal is to finish it. And so how can you redirect all that energy?

I thought it was really cool. I was going through that post that you did on LinkedIn, where it's basically saying, "hey, you're oversubscribed at this point. You say three is your max and you've got six projects going on. You need to tone this down."

Yeah, and I think those are the types of things that AI is just amazing for. Just to give you a quick kick in the ass and be like, "hey, donkey, you said that you don't want to do this, why are you doing this?"

That's great. So how do they get to Ace?

It's not — yet you can sign up for it. It won't be very useful for you. As of a week ago, I have published the CLI to npm that goes with it, that you can use inside of Claude Code. It OAuths into the app, and Claude can — so you have an MCP. It's just a CLI posted. But I've got a public repository with a CLAUDE.md and a bunch of skills in it that are meant to teach Claude how to use the Ace CLI to go back and forth. It can do anything that can be done inside the app — it can create projects, it can do all that sort of stuff.

But also, it is integrated with everything that I use — from Linear to GitHub to Slack. It'll pull in all the activity that's happening in there. And I can feed that to the assistant to say, "does this align with what your projects are? Maybe you need to rearrange those."

That's awesome. That's absolutely awesome. There's that. And I also, as a fun thing, started a new Instagram called Slop of the Day.

I gotta look this up. So for RootNote, I constantly need test content. And I was using my own stuff, but I'm too precious about my own stuff. It has to be good if I'm posting it. So I started this account that I can just literally — whatever's on the top of my head, I'm going to generate it in Vo or in Sora or whatever else and post it as an Instagram reel or post, so that I always have some test content to do. But I started this literally yesterday.

All right, cool. There you go. There's your slop source of your content. Love it.

Well, thanks guys. Craig, thanks for coming here and chatting with us. And Brandon, as always, thank you.

I'll check — we'll check out your OnlyFans page.

Oh yeah, wait — that's right. So I'm going to be starting an OnlyFans page. And it's going to be me coding nude. And if you're interested in short little hairy leprechaun-looking guys coding, you let me know and we'll do it.

We'll just cut that out. This has been — oh, by the way, I'm editing, so no, that's not going to be cut out.

All right, there's something out there for everyone.

Yeah, thanks guys. This has been the Big Cheese podcast. We'll see you next time. Peace.