The Friday afternoon list just got shorter
The new language models aren't making my main work faster. They're making the little jobs I always skipped cheap enough to finally try.
Every team I’ve ever seen keeps a secret list. It’s the stuff too small for a roadmap and too big for a Friday afternoon. A script to clean out stale feature flags, a dashboard for the query someone runs by hand every Monday, a quick prototype to settle an argument that has been going in circles for three weeks. Those things mostly never happen. They sit there like a drawer of dead batteries you keep meaning to recycle.
GPT-4 came out a couple of weeks ago, and I’ve been poking at it in the evenings. It still makes things up, in the same calm voice it uses when it’s right. Where it surprises me is starting. Something I’d have waved off as a day of wrestling an unfamiliar library turns into an hour of describing what I want and fixing what it got wrong. It doesn’t have to be right. It just has to hand me something I can check faster than I could have written it.
So the bar for starting dropped through the floor, and the bar for shipping didn’t move an inch. Maybe the drawer finally gets emptied. Maybe it just gets fuller, because now every idea feels cheap enough to toss in.