GPT-5 Pro Ate 1300 Lines and Asked for More
The file I'd been avoiding for months is gone in one shot.
GPT-5 Pro ate 1300 lines and the file survived. Large-scale AI refactoring became routine. The file you dread opening now has someone willing to open it for you.
3200 lines down to 1900 in a single pass. No functionality lost. No hand-holding. I handed it the file and it just — went.
That file had been on my mental list for months. The kind of thing where you open it, feel a small death inside, and close it again. I was dreading the day I'd have to sit with it as an actual human and make decisions about every redundant abstraction and every copy-pasted block of shame.
GPT-5 Pro didn't dread it. It demolished it.
I've been using Sonnet 3.5 as my cleanup pass guy — perfectly good at what it does — but this was a different category of task. The biggest refactor I've thrown at any model. And the thing that got me wasn't that it succeeded, it was that it succeeded in one go. No chunking, no "here's part one of three," no asking me what I wanted to keep. It just understood the shape of the problem and collapsed it.
Now I'm going through every offending file in the codebase. One by one. The hits keep coming.
There's something quietly disorienting about this — not scary, not exciting, just slightly surreal — the way a piece of work you'd been dreading stops existing on your list because a model ate it for breakfast. It wasn't even the hardest thing it did today. It was a warmup.
Counterpoints
Push back, extend the argument, or sharpen it. New counterpoints go through review before they show up here.
No approved counterpoints yet.