Not enough sleep, too much AI in my head

I’ve been using AI tools for coding for a while now. GitHub Copilot Pro, the occasional ChatGPT session — I’ve been paying for this stuff and watching it closely. But I recently had what I can only call my actual AI realization moment.

Two things clicked, roughly at the same time.

The first: the hype was never wrong, just early. For years I nodded along to “AI will change everything” while quietly thinking: yeah, but not quite yet. The tools were useful, not transformative. What changed isn’t one model or one provider — I use Claude, I’ve used Gemini, I’ve used Copilot — it’s that the combination of significantly better models and dramatically better tooling has finally crossed a threshold. The gap between “impressive demo” and “actually does the work” closed.

The second: I stopped reviewing every diff. With Copilot I always had VS Code open, manually accepting changes one by one. With Claude Code I don’t necessarily have an editor open at all. That shift sounds small. It isn’t. The reason it happened isn’t blind trust — it’s that when I properly prepare the context upfront, think through the problem before I type anything, and ask the right questions first, the output is genuinely close to what I would have written myself. The control step stopped being necessary because the quality made it redundant.

I’m using this mainly for Dartobert, my darts tracking app. On day two I upgraded from the €20 plan to the €100 plan. That says everything about where my head was at.

Here’s the honest part: I’m sleeping less than I should. I’m up late every night because I can’t stop. There’s this irrational feeling that all 100 projects I’ve had on my list for 20 years can now be finished — simultaneously, this week. I know that’s absurd. But Dartobert is actually moving forward, my website is getting updates, and I’ve got three other things running in parallel that I haven’t touched in years.

Something shifted. I’m pretty sure I’m not going back.