It Works. That's the Problem.

I've been skeptical about AI for a while. Not "shut it all down," but the kind of skeptical where you follow incentives to their logical conclusion and ask how this all ends.

I never saw it ending well.

So naturally, like any developer, I went and played with it. And here's the uncomfortable part: It works.

For code especially, it works shockingly well. I haven't written a line of code in about five months. Our biggest bottleneck right now isn't bugs or outages … it's PR throughput. We're reviewing more changes than we can comfortably process. The output is dramatically higher than it was before. Estimating a ticket? You write the ticket and the work simply materializes.

The kanban board may be the new editor.

We're a small startup. Efficiency is good. Big thumbs up all around! The code quality hasn't collapsed. We're heavy TDD, strong domain boundaries, reasonably clean architecture. In a well-structured codebase, the models are legitimately productive.

That's the part that's hard to dismiss as hype.

You might be thinking, "Cory, I tried this, its dogshit!" My friend, I have unfortunate news, it isn't anymore.

My biggest fear used to be the economic fallout when this bubble pops. Now it's that if it pops, I have to go back to the way I used to develop.

The thing I used to love doing.

People worry about the code being sloppy. That's not my experience. I can give it brutal, specific feedback and it just adjusts. No ego. No pushback. You can act like Linus! It's fine! Code quality is better. The product is better. The customers are happier. We're more stable than we were before. This has unleashed a nitpicky, polished, exactly-the-way-I-want-it-to-work kind of output that I never want to lose.

And that's also the part that's making me uneasy.

Because while we're all celebrating leverage, we're also quietly building economic dependence on a very small number of companies.

If intelligence, interpretation, synthesis, code generation, and automation become a rented utility controlled by 3–5 providers, that's not just another tooling shift. That's structural leverage over the economy.

Startups build on it. Enterprises rely on it. Governments integrate it. Once that layer hardens, it's very hard to unwind.

We've already seen what happens when foundational internet layers consolidate. Social platforms centralized distribution. Journalism economics broke. The open web eroded. Ad. Video. Clickbait. 100 words describing a one sentence tweet.

Now we're centralizing interpretation itself … the interface to knowledge and decision-making. The open web is toast. Why is anyone even incentivized to build a web page anymore? Google will just summarize the results.

And most of the loud conversation is about speed. How much faster we can ship. How much leverage we can extract. How much competitive advantage we can gain.

All of which is real. I'm benefiting from it too.

But I'm surprised how little serious discussion there is about long-term concentration and dependency, especially from people who built their reputations talking about decentralization, open systems, bias, access, power dynamics, responsible innovation.

If you believe those things matter, this layer sits underneath all of them.

I'm not saying "stop." I'm not pretending I'm not using the tools. I am.

I'm saying the tradeoffs here are enormous, and we're moving very fast. Maybe open models win, but good luck getting the capital to buy GPUs. Maybe competition fragments the space. Maybe this consolidates even further. And by the time the skeptics admit it works, it'll be too late to renegotiate the terms.

I don't know.

I just know it feels strange to watch the intelligence layer of the global economy consolidate this quickly while most of us are understandably distracted by how productive it makes us.