First test
That framework migration that’s been sitting on the backlog for a year. Multiple projects, library consolidation, toolchain upgrades, devops config. The kind of thing you keep pushing because it touches everything.
It took an afternoon. Toolchain, dependencies, data layer, tests – done. Edge cases I usually catch on the second pass, handled.
I knew the models were getting better. I didn’t realise the gap had closed that much.
Second test
I wanted to see if a model could do the planning part. Architecture decisions, figuring out the order of operations, catching edge cases upfront. The bit that usually takes me the first half of a project.
Gave it a spec for something new. It designed the structure, sequenced the work. I reviewed, tweaked a few things, and spent the rest of the day directing execution. By evening it was deployed, documented, tutorialed, with walkthrough videos and voice-over.
I spent zero time on the part I used to spend the most time on.
What I keep coming back to
You still need to know what you’re doing. If you can’t tell good output from bad, the models won’t fix that. But if you can, the work feels different now. Less building, more steering.
Not dramatic. Just different.
I have favourites like everyone. The floor has risen. What was a stretch six months ago is routine, and what’s a stretch today will be routine in a few months.
Things move that fast now.
I went looking for where the models break. Found less than I thought I would. The interesting questions keep shifting. Curious where they land next.
답글 남기기