OpenAI Codex Improves Rapidly, Rendering User Tricks Obsolete
I used to have this sort of "meta repo" structure for all repos and projects I would work on, with one repo just for agents where I would force them to write progress logs, mention bugs, put shared tools for vms or debugging.
Now they don't seem to need any of that.
I feel like codex is getting better so fast that my tricks I used to do to make it more productive are outliving their usefulness faster than I can come up with them.
In fact, I have found that most of the time I need notes on what they have learned, rather than the other way around. I find myself asking them to keep notes up to date in Notion so they have more context about what the team is doing to inform and motivate what I'm building.
I used to have this sort of "meta repo" structure for all repos and projects I would work on, with one repo just for agents where I would force them to write progress logs, mention bugs, put shared tools for vms or debugging. Now they don't seem to need any of that.