The question is if I am becoming better at thinking and planning at the expense of losing the ability to turn words into code or if this is a larger kind of regression that I am experiencing that is making my overall capabilities drop.
Although I am becoming faster and better at using AI I can keep more agents busy with less effort, plan larger scopes and keep more moving parts in my brain which is completely different than what I used to do which was keeping the entire source-code in my brain. I can still recall pretty much every project I ever interacted with and tell you exactly how any part of it works or the vague code flow which made me really fast at figuring out bugs and fixing them. Of course I cannot do that anymore with new projects and rely on AI to figure bugs out.
The kicker here is that instead of knowing how an application works from knowing the sourcecode instead I know how an application from the abstract user-interaction flow which can develop inconsistencies since there's no ground truth anymore and that part worries me, but I do maintain a strong documentation now - something I never used to do since I have the unique capability of recalling the entire project I am working on - which ironically also improves agent performance since they have something to work from on fresh sessions instead of reinventing the wheel and having to re-explore.
Only time will tell if this is a disaster in the making, but I am at least 3 times faster than I would have been without AI. In some cases I cut down work that used to take me weeks to around a day, but that's mostly UI related that never needed any critical thinking to begin with.
That changes the failure mode. Before, the bottleneck was speed—you understood everything but couldn't build fast enough. Now it's judgment—you can build fast but you can't always tell when something's wrong.
The documentation habit is you compensating—building external ground truth to replace the internal ground truth you lost. Which is not regression, but it's also not nothing. A calculator offloads computation. This offloads the knowledge that tells you when the computation is wrong. Different thing.
Uh..... Do you know about AI Agents? Claude Opus?
Dont judge AI on the instant chatGPT model.
This quote from Robert A. Heinlein comes to mind
Which I'm pretty sure is the plan, FWIW.
I'd be okay if we treated screen time and social media like cigarettes. We can talk about AI after.