@rachel ill, but I’m not convinced of melodramatic. I’m having similar thoughts of late, I often wish I could be one of the “ugh, it doesn’t work and nobody asked for it” people because it seems much easier than being a “it works really well, can be useful tool, but using it requires funding some of the worst people around” person. Also as you mention, spending all day doing code review for a machine just isn’t as fun as sitting down and thinking hard about how to do things.
@jon I actually think that the code review problem will largely solve itself, we will get fed up and just stop, and keep making agents better at it. Human nature will override good sense and we will rely on a few good people to keep making it better.
@rachel this immediately brought to mind Warhammer 40,000 with its tech priests who worship the machine god.
Not reviewing the code seems even worse than being stuck doing so, because the next step is writing endless spec documents that have to be reviewed to make sure everything has been covered to a degree the machine can’t possibly miss implementing it.
@jon that does seem like a potential and rather tedious outcome. However a product manager can back and forth on an implementation on pen and paper, or rapid development. So too can an agent. I am sure we'll develop systems to help with this. And of course it is going to differ project to project, and on the size of the codebase. I wonder if micro-services will make more sense again vs monoliths for small teams.
@jon btw I have been playing with Cucumber again, so there is a spec that describes how a user interacts, and it can't go out of sync because it is the tests.
@jon I have been running an experiment where I lean in on DHHs rails world keynote and see what it would take for me to be happy merging LLM code without reviewing it. It changes the way you think about it.
* I am not merging code anywhere important in this experiment.