This is something else entirely. I've been working on a long-form piece on using AI in a large, in-production codebase for the last month or so. My tack has been towards caution and towards measured usage of Claude 4.5 and GPT 5.2 in specific, targeted aspects of the SDLC.

The last week working with 4.6 has been eye-opening. It's completing and even occasionally one-shotting tasks - including coding - that previously I would expect to have to have a long, sometimes painful, sometimes slower than just doing it myself iteration with the agents.

Not saying this to scare anyone or create FOMO. More like to put a bookmark in time - here is where I felt a shift, a passing of the tipping point of utility. A moment when before this I could say that using AI as a software developer was a very good thing to explore tactically, and after which I may end up saying that not using AI as a software developer makes little sense strategically.

I'll be back in a few weeks with that more thorough post...

Edit: You can find it here.