Post

Three weeks ago, I joined an internal AI hackathon. The rule: one day, build anything work-related — but rely almost entirely on AI. Obviously, I chose to refactor a core, complex part of the system. You know me — I like a bit of #DDDesign chaos. A thread 👇 1/9

If someone says AI writes all their code, one of two things is true: the problem is extremely simple or they’re overselling it In real systems, with real constraints, you still need actual engineering. 2/9

For small changes, it’s helpful. AI picks up some context and gives some good answers. But it always needs review. Always. And even then, small bugs slip through. 3/9

For complex logic? It gets slippery. You correct mistakes → update the prompt → things look good… Then the same bugs sneak back in. Even if you explicitly told it not to do that. Frustrating. 4/9

That said — it’s not useless. I used it to explore ideas, unblock some stuck thinking, and even got a few design insights I hadn’t considered before. It’s a good brainstorming partner. 6/9

But when it came time to implement those insights? Nope. It couldn’t follow through. The gap between idea and reliable code was just too wide. Prompting alone couldn’t bridge it. 7/9

I still think it’ll improve. Models will get faster. More accurate. Maybe we’ll trust them more. Or maybe we’ll just get better at supervising more changes — safely. 8/9