What AI Actually Did to My Work in 2025 — An Honest Recap

Somewhere around March this year I stopped feeling like I was experimenting with AI and started feeling like I couldn't work without it. That's a bigger shift than I expected.

I want to be honest about what that actually means in practice — because most of the takes you'll read are either breathlessly enthusiastic or wilfully cynical, and neither is particularly useful if you're trying to figure out whether any of this matters for your business.

The thing that actually changed

The most significant shift wasn't any specific tool. It was that the boring, repetitive parts of building software stopped costing real time.

Every project has grunt work. Scaffolding a data model. Writing a form with fifteen fields and the same validation logic on each one. Setting up authentication for the fourteenth time. Wiring up a payment provider using an API you've used before but can never quite remember the shape of. I'd always factored this into estimates because, well, it takes time. It just does.

In 2025, it mostly stopped taking time. Not because AI does it perfectly — it doesn't — but because AI does a good enough first pass that the remaining work is editing, not building from scratch. That's a meaningful difference. Editing is fast. Editing is also where your actual judgment matters.

The result was projects that came in faster and cheaper than the estimates suggested. Which is a nice problem to have.

What I was wrong about

I thought the biggest win would be the clever stuff. Architecture suggestions. Debugging complex problems. Code review. And those things are genuinely useful — Claude catching a race condition I'd have spent an hour finding manually is a good day. But it's not where the time savings are. The time savings are in the tedium.

I also thought I'd be able to delegate more completely. I can't. Every meaningful decision still needs a human — the AI doesn't understand the client's weird edge case, or the fact that performance matters more than elegance here, or that we can't change the database schema without a migration plan. The tools have gotten better at pointing in the right direction, but they're not running the project.

What was mostly hype

"AI will write your entire app." It won't. Or rather: it will produce something that looks like an entire app and then you'll spend three times as long as the original build would have taken trying to fix the parts that are subtly wrong.

The closer the task is to "something I've seen before, well-defined, relatively isolated" the better AI is at it. The further it gets from that — novel product decisions, complex integrations, anything involving your specific business logic — the more you need a developer who actually understands what they're building.

The best use of these tools is removing friction from the known parts so the humans have more capacity for the unknown parts. That's less exciting than the pitch, but it's actually quite valuable.

What I'm carrying into 2026

Honestly, I'm more bullish on AI tooling than I was in January — not because the hype was right, but because the practical reality has settled into something genuinely useful. The tools are getting better. The workflows are clearer. I know which tasks to hand off and which to keep hold of.

What I'm less certain about is what happens as the tools continue to improve. At some point the calculus changes. I don't know exactly when, but it feels like something worth paying attention to.

If you're a business owner wondering whether this stuff matters for your next project: it does. Not in the dramatic way the headlines suggest — no robots taking over, no magic buttons. In the practical way of projects that come in a bit faster, a bit cheaper, with a bit more of the developer's attention on the parts that actually need it. That's a less exciting story. It's also a real one.