I largely agree with the gist of this article but its calculation about productivity is very flawed as it doesn’t account for the time things sat on the backlog, or the things that wouldn’t have been done at all.
Where I see major productivity gains are on small, tech debt like tasks, that I could not justify before. Things that I can start with an async agent, let sit until I’ve got some downtime on my main tasks (the ones that involve all that coordination). Then I can take the time to clean them up and shepherd them through.
The very best case of these are things where I can move a class of problem from manually verified to automatically verified as that kick starts a virtuous cycle that makes the ai system more productive.
But many of them are boring refactors that are just beyond what a traditional refactoring tool can do.
It's allowed me to be even more of a perfectionist. I'm quite enjoying it, and I review and revise everything that's produced line by line.
I doubt that's the commonly desired outcome, but it is what I want! If AI gets too expensive overnight (say 100x), then I'll be able to keep chugging along. I would miss it (claude-code), but I'm betting that by then a second tier AI would fit my process nearly as well.
I think the same class of programmers that yak shave about their editor, will also yak shave about their AI. For me, it's just augmenting how I like to work, which is probably different than most other people like to work. IMO just make it fit your personal work style... although I guess that's problematic for a large team... look, even more reasons not to have a large team!
Where I see major productivity gains are on small, tech debt like tasks, that I could not justify before. Things that I can start with an async agent, let sit until I’ve got some downtime on my main tasks (the ones that involve all that coordination). Then I can take the time to clean them up and shepherd them through.
The very best case of these are things where I can move a class of problem from manually verified to automatically verified as that kick starts a virtuous cycle that makes the ai system more productive.
But many of them are boring refactors that are just beyond what a traditional refactoring tool can do.