dax: your custom agent setup solves problems that are gone
dax says the models now improve faster than the people tuning them, so someone naively running vanilla codex is more likely to be at the state of the art than you with your rules files. Kent C. Dodds agreed and went further.
The big disagreement
dax says custom agent setups fix problems that no longer exist
dax argues that LLMs improve faster than the tinkerers, so elaborate custom workflows are mostly patches for model weaknesses that have already been fixed. Kent C. Dodds agreed and added his own rule: make your setup as vanilla as possible, and give the agent tools it can use rather than text instructions it must religiously follow.
the person naively using vanilla codex is more likely to be experiencing state of the art
Receipts
Theo fixed a GitHub annoyance with a 90 second Chrome extension
Theo asked Opus 5.5 to fix one of his biggest GitHub annoyances, clicking screenshots opening them in a new tab, and says he had a working Chrome extension 1.5 minutes later. He concedes it is basic stuff. His point is that the cost of fixing small irritants in software you do not own has collapsed.
I know this is basic stuff, but it still feels magical that you can just fix things.
How they actually work
Pocock's prompt of the day rewrites your AGENTS.md three ways
Matt Pocock posted a prompt that asks the agent to restructure a messy AGENTS.md: remove no-ops, use progressive disclosure, and move instructions into CODING_STANDARDS.md. The trick is the last line, which tells the agent to apply the work over three subagents, each more radical than the last, and open a single PR.