Latest / The Bootstrapped Tech Founder with Fexingo: Profitable Software Companies Without VC / Why Bootstrapped Founders Build In-House AI Tools
Transcript
- Lucas: If these conversations are useful for what you're building or running, we want to be upfront about how this show exists. A handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that's literally what funds making this many of these. No ads, just you keeping the lights on. Luna: It’s amazing how much consistency that small group buys us, especially when the algorithm changes hit so hard. Lucas: Right. And it keeps us from having to chase clickbait headlines just to pay the servers. So let’s get into today’s angle. For years, the advice for bootstrapped tech founders has been 'don't build AI.' It’s too expensive, too complex, and you should just bolt on a generic chatbot widget. But I’ve been looking at the recent moves by some highly profitable, independent software companies, and they’re doing something radically different. Luna: You mean they’re actually building their own AI infrastructure instead of just using OpenAI’s API? Lucas: Exactly. They’re not building foundational models from scratch, which would be insane. They’re building specialized, in-house AI layers that solve their own internal problems first. Think about a company like Notion or Linear. They didn’t wait for investors to tell them to add AI features. They built internal tools to help their support teams triage tickets faster, or to auto-summarize user feedback. That internal efficiency became the foundation for their public-facing features. Luna: That makes sense because their data is already siloed in their own database, so training a model on their own usage patterns is cleaner. Lucas: It’s cleaner and cheaper. Here’s the concrete math most people miss. If you’re paying per token for every user interaction with a third-party API, your costs scale linearly with your growth. But if you build a lightweight, specialized model or a sophisticated rule-based automation layer in-house, your marginal cost drops significantly after the initial build. You’re trading capital expenditure for operating expense, but over time, the OpEx savings compound. Luna: So it’s basically buying equity in your own technology stack instead of renting it from Big Tech. Lucas: Precisely. And there’s a second benefit that doesn’t show up on a P&L statement immediately: data privacy and security. When you use a third-party API, you’re often sending customer data out of your control. For enterprise clients, that’s a dealbreaker. By keeping the inference or the logic layer internal, you can offer stronger privacy guarantees. That becomes a sales feature, not just a compliance checkbox. Luna: But doesn’t hiring engineers to maintain that custom AI layer create a new kind of technical debt? Lucas: It does, if you do it wrong. The mistake bootstrapped founders make is trying to replicate what Google or Meta spent billions on. You don’t need a massive neural network. You need a focused application of existing open-source models or even deterministic logic that mimics AI behavior. The key is starting small. Build the tool for yourself first. If it saves your team ten hours a week, then it’s worth five hundred dollars a month in server costs. Then you can productize it. Luna: So the metric isn’t ‘how smart is our AI,’ it’s ‘how much time did it save our ops team?’ Lucas: That’s the golden rule. Look at how some smaller CRM platforms are handling lead scoring. Instead of paying for a cloud-based prediction service, they’ve built simple decision trees based on their own historical conversion data. It’s not magic, but it’s accurate enough for their niche, and it costs them almost nothing to run. That’s sustainable growth. Luna: It feels like the opposite of the race to the bottom on price per feature. They’re raising the floor on quality. Lucas: They are. And it creates a moat. Competitors who are relying entirely on third-party APIs can copy your feature next week. But they can’t easily copy your proprietary data pipeline or your internal workflow integrations. That’s unfair advantage number one hundred and eight. Luna: I’m curious about the talent aspect though. Do you need a dedicated ML engineer, or can generalists handle this? Lucas: For the early stage, generalists can absolutely handle it. Most modern developers are familiar with Python and basic vector databases. You don’t need a PhD. You need someone who understands your business logic deeply. That person is usually already on your team, maybe in product or engineering leadership. They know where the friction is. Luna: So it’s less about hiring a new role and more about empowering existing roles to think differently about automation. Lucas: Spot on. And here’s the counter-intuitive part. By building in-house, you force yourself to understand your own product better. You can’t automate what you don’t understand. So the process of building the AI tool often reveals inefficiencies in your core product that you’d otherwise ignore. It’s a diagnostic tool as much as it is a feature. Luna: That’s a beautiful side effect. You fix the leaky bucket while you’re installing the hose. Lucas: Exactly. Now, let’s talk about the risks. The biggest risk is over-engineering. It’s easy to spend six months building a sophisticated recommendation engine that only improves conversion by two percent. For a bootstrapped founder, that’s wasted runway. You have to tie every hour of development to a measurable outcome. Either it saves money, increases retention, or enables a higher price point. Luna: So if it doesn’t move the needle on those three metrics, you scrap it and go back to manual processes. Lucas: Yes. Manual processes are underrated. Sometimes the best AI is just a well-written email template that gets copied and pasted. Don’t build a robot to do what you can do in thirty seconds. Only build the tool when the volume makes human execution impossible or prohibitively expensive. Luna: Volume is the trigger. Without volume, automation is just procrastination. Lucas: Love that. Another consideration is vendor lock-in. If you build your entire business logic on top of a single third-party API, and they change their pricing or shut down, you’re dead in the water. In-house tools give you optionality. You can swap out the underlying model or the library without rewriting your whole application. That resilience is priceless. Luna: It’s like owning your home versus renting. Renting is easier until the landlord raises the rent or sells the property. Lucas: Or evicts you. Right. And speaking of ownership, there’s the cultural impact. When your team sees you investing in their workflows, morale goes up. Engineers love building cool stuff. If you let them build internal tools that make their jobs easier, you retain top talent better than any bonus package. That’s a soft benefit, but it’s huge for long-term stability. Luna: So it’s a retention strategy disguised as a productivity hack. Lucas: It is. And finally, think about the narrative. Customers are tired of feeling like data cattle. When you can say, 'Our AI runs locally on your instance' or 'We don’t train our models on your data,' you stand out in a crowded market. That’s a powerful brand signal for privacy-conscious enterprises. Luna: Trust is becoming the scarcest resource in software. Owning your stack builds trust. Lucas: Absolutely. So, if you’re a bootstrapped founder listening to this, ask yourself: what repetitive task is eating your team’s brain space? Is there a way to automate it internally before you try to sell it externally? Start there. Build the shovel before you dig the mine. Luna: Build the shovel first. Because if you don’t have a good shovel, digging might just break your back. Lucas: Right. And once you have that shovel, you can dig deeper than anyone else. It’s not about being the fastest; it’s about being the most durable. That’s how you build a company that lasts decades, not just quarters. Luna: Durable over fast. That’s the bootstrapper’s mantra. Lucas: It is. And as the market shifts in late twenty twenty-six, with interest rates stabilizing and capital becoming slightly more available, the pressure to raise VC will return. But remember, you don’t need outside money to build intelligent systems. You just need clarity on what problems are worth solving. Luna: Clarity over capital. Always. Lucas: Exactly. So here’s my question for you: what’s one internal bottleneck in your business that you’ve been ignoring because it seemed too small to automate? What if you spent a weekend fixing it? Luna: I bet the answer surprises you, and it probably doesn’t require a PhD to solve. Lucas: Maybe not. But it might just save your sanity. Thanks for listening.