I want to talk about something that's been on my mind for about eighteen months now, and I keep seeing founders and operators miss it: the economics of running a software company are genuinely different now, and most people are still playing by the old rules.
This isn't hype. It's arithmetic.
The Cost Structure Broke
Three years ago, if you wanted to build a product seriously, you needed a team. A real one. Designers, frontend engineers, backend engineers, infrastructure people, QA — minimum viable headcount to ship something that didn't fall apart. That meant burn. That meant runway. That meant investors.
Today, a single person with access to the right tools can move at a pace that would have required a team of eight two years ago. I'm not talking about prototype-quality work. I'm talking production software. I've watched this happen from the inside.
The implications are enormous and most people haven't followed the logic through.
When your cost to build drops by an order of magnitude, everything downstream changes: how much capital you need, what kind of capital makes sense, what "traction" looks like before you raise, whether you even need to raise at all.
You Are Arbitraging Billions
Here's the part that feels almost unfair: the big AI labs are spending billions of dollars competing for your usage. OpenAI, Anthropic, Google, Meta — they are subsidizing your access to capabilities that would have been impossible for a small team to build even at any cost two years ago.
Every dollar they spend training frontier models and making them accessible via API is effectively a subsidy to anyone building on top of them. You're using compute that cost hundreds of millions to produce, for pennies per call.
The arbitrage is: they fund the infrastructure, you capture the value in the application layer.
If you're not actively thinking about how to maximize that subsidy — what you can build because of it that you couldn't have built before — you're leaving the most significant wealth transfer in software history on the table.
The window won't stay this wide forever. Models get commoditized, margins compress. But right now? The cost-to-value ratio is extraordinary.
What "Magic" Actually Costs Now
Let me give you a concrete example of how I think about this.
Two years ago, a product that could reliably read a complex document, extract structured information, reason about it, and surface a useful action — that would have required a PhD-level ML team, months of training data curation, fine-tuning infrastructure, and significant ongoing model maintenance. Call it $500k to get something production-viable, and another $200k/year to maintain it.
Today, you can prototype that in an afternoon. Get it production-ready in a week. Run it for single-digit cents per operation.
The things that used to "feel like magic" — because only Google or Stripe could afford to build them — are now buildable by anyone who knows how to wire an API together and has good product instincts.
The question shifts from "can we afford to build this?" to "are we thinking ambitious enough about what's possible?"
You Don't Have to Be Venture-Funded
This is the thing I keep saying that surprises people most.
The old equation was: ambitious software company → need capital → need investors → need venture-scale returns → need to optimize for a specific exit path. The entire logic chain was forced by the cost structure.
When your cost to build collapses, the revenue you need to sustain and grow also collapses. A product with 1,000 paying customers at $50/month is $600k ARR. With a lean team and AI-native tooling, that's a healthy, growing company. Three years ago, that same product might have required $2M+ in infrastructure and headcount to operate.
That doesn't mean VC is wrong. For some businesses, at some stages, it's exactly right. But it should be a deliberate choice, not the default path because you couldn't afford to move without it.
I want to build things that generate real revenue from real users — not because I'm allergic to investment, but because a business that makes money from its customers has a fundamentally different relationship with reality than one that's funded on projections.
Both can be great. But the choice is yours now in a way it wasn't before.
The 1-3-10 Framework for Users
One thing that has clarified my thinking is what I call the 1-3-10 framework for getting users.
The first user is always yourself, or someone you know extremely well. You're not building for the market yet — you're building to solve a real problem that you or someone close to you actually has. This is your proof that the problem exists and that you can solve it.
The next three users validate that other people have the same problem. They're usually found through direct outreach, your network, or communities you're genuinely part of. Don't optimize this stage. Manually find them. Talk to them. The goal isn't scale — it's signal.
Ten users is where you start understanding if you have something repeatable. Can you find ten people who aren't connected to you personally and get them to pay? If yes, you have early evidence that the market is real.
The mistake I see constantly: people jump from 1 to 10,000 in their heads before they've done 1 to 3. They build for scale before they've validated signal. With AI tooling, the temptation is worse because you can move so fast — but moving fast in the wrong direction is just efficient failure.
Go get your first user. Then three. Then ten. Make money at each step.
The Building Blocks Point
Here's the last thing I want to say, because it's the one that requires the longest time horizon to see.
The products we're building at Derivative Labs aren't isolated bets. Each one is a building block in a larger system. The infrastructure for one creates distribution for another. The users from one create a market for the next.
This is only possible because we're not spending all our capital on headcount — which means we can afford to think in years instead of quarters. The low cost structure buys us something more valuable than time: it buys us patience.
The arbitrage isn't just financial. The real arbitrage is that AI tooling lets a small, focused team think and operate more like a long-term compounder than a sprint-to-exit startup.
The economics changed. The way you build and capitalize should change too.
This is part of our Building in Public series on Applied AI. I'm writing these to share what we're actually learning, not what sounds good in a pitch deck.