I've been watching a few AI rollouts recently. Copilot enabled across the org. Usage through the roof. Engineers loving it. And yet... nothing actually improving.

Cycle time? Same. Bugs in prod? If anything, slightly worse. What's going on?

I've been thinking about this a lot.

We've been here before. Sort of.

Every few years there's a new wave. Agile. DevOps. Microservices. Cloud. Each time there's hype, then reality, then we all adapt and move on. Most of us have survived multiple waves by now.

So when people tell me "AI is just another hype cycle, we'll figure it out" I get it. That's been true before.

But I'm not sure it's true this time. Here's what's bugging me.

All those previous waves changed how we work. New practices, new tools, different ways of organising. But one thing stayed constant: human speed. A developer writes code at roughly the same pace whether you're doing waterfall or agile or whatever.

AI breaks that. Suddenly code gets generated way faster than before. And that sounds great until you realise everything else stays the same speed.

Reviews? Same capacity. Testing? Same discipline. Deployment? Same pipeline. Learning? Same brains.

So what happens when you pour 3x more code into a system that can only review at 1x speed? You get a pile-up. PRs waiting. Reviews getting rushed. Things slipping through.

That's exactly what I've been seeing. Teams generating more code than ever. But the review bottleneck means stuff is either waiting forever or getting rubber-stamped. Neither is good.

The thing that keeps me up at night

Here's what really worries me though.

When you speed up feedback loops, you don't just go faster. You converge faster. On whatever your system naturally tends toward.

Good foundations? Fast loops help you get to good solutions quicker.

Shaky foundations? Fast loops help you converge on your existing dysfunction. Also quicker.

Most leaders I talk to are watching outputs. Lines of code. Features shipped. Velocity metrics. They're not watching what those outputs are converging toward. By the time the problems show up in the output, you've already run the loop hundreds of times.

What I've started asking

When I talk to engineering leaders about AI adoption now, I've started asking different questions:

If your team's code output doubled tomorrow, what breaks first? Is it reviews? Testing? Deployment? That's your actual constraint, and AI is about to stress-test it.

What is AI actually accelerating in your org right now? Is it good outcomes, or is it amplifying problems you already had?

The tech isn't the hard part. Being honest about whether your org is ready to go faster... that's the hard part.

Curious if others are seeing the same thing. Hit me up on X if you've got a different take.