AI-Assisted Code Isn't the Problem. Inexperience Is.

There's a narrative forming that AI-written code is sloppy, insecure, and dangerous.

I get where it comes from. If you hand a tool to someone who doesn't know what good looks like, you get bad output faster. That's true of AI. It's also true of Google, Stack Overflow, and every other tool we've ever used to build software.

But I want to make a claim that might bother some people:

AI-assisted code, built by an experienced developer working alone, can be more secure than code written by a team.

Not always. Not in every context. But more often than the "AI slop" crowd wants to admit.

Here's why.

A team has ego. A team has deadlines that make people merge PRs they know aren't ready. A team has "that's not my part of the codebase" blind spots. A team has the Friday afternoon deploy that nobody wants to be the one to block.

AI has none of that.

When I'm building alone with AI, every line of code gets reviewed by something that doesn't get tired, doesn't have politics, and never decides an edge case isn't worth thinking about.

That doesn't mean AI is always right. It's not.

A few weeks ago I was building a side project — a fast Bitcoin prediction game called Binary — and AI confidently suggested an approach to handling concurrent wagers that would have introduced a subtle race condition. It looked clean. It passed every obvious test. And it was wrong.

But here's the thing: I caught it. Because I've shipped enough code to know what breaks under load. AI doesn't replace that instinct. Nothing does.

And that's exactly the point the discourse keeps missing.

The risk isn't that experienced builders will ship worse code with AI. The risk is that people who've never shipped anything will produce code that looks finished but isn't. Those are two completely different problems, and conflating them is going to cost people who sit on the sidelines while the rest of us build.

Binary is a good example of what I mean. I'd been thinking about it for months — not as "I should build a gambling app," but as a fun engineering challenge I couldn't shake. I started by writing a couple test bots as a curiosity project: could I pressure-test the system, and what breaks when you actually try?

The conclusion was pretty clear: this style of trading isn't my thing. But the process was useful because it exposed how rough the underlying tech really is. Things were slow, fragile, and full of gotchas that only show up once you try to automate anything.

That's what made it stick. Not the "money" part — the infrastructure. So I treated it like a constraint puzzle: could I build a fast, simple 15-minute BTC prediction game that settles natively on Bitcoin and Lightning — no bridging friction, no duct tape, just quick rounds and clean settlement on the network the game is actually about?

A few months ago, this would have taken a team and a quarter to even explore. Instead I prototyped, pressure-tested, and shipped a testnet in weeks. Not because AI did the work — because AI removed the friction between the idea and the thing.

I wrote a few weeks ago about the person you're about to become. About how AI changed my ambition and made me curious about building again after years of coasting.

Binary is what happens when that curiosity meets less friction. An idea that was too heavy to justify suddenly becomes light enough to try.

The code isn't sloppier. It's tighter. The security isn't weaker. It's stronger. Not because AI is magic — but because a tireless collaborator that never gets bored of reviewing your edge cases makes you a better builder. And when it's wrong, you're still the one who decides.

The people shouting about AI slop aren't wrong about the symptom. They're wrong about the cause.

The problem was never the tool. It's always been the builder.

Testnet is live if you want to try it: app.binary.fun — 15-minute BTC predictions, settled on Lightning.

Honest question: do you trust code that was built by one experienced developer working closely with AI, or code that went through a team's review process on a Thursday afternoon before a long weekend?

I know my answer. Curious about yours.