been juggling cursor + openai codex this week
CURSOR (with gpt-5) = power drill for messy multi-file refactors
CODEX = the robot intern for tests/chores 😅
tricks
-> keep asks tiny (one diff at a time)
-> be super explicit (file paths + “done-when”)
-> ctrl+I opens the agent, ctrl+E shows background agents
-> let codex run in its sandbox while you keep typing
-> add a tiny agents.md so both stop guessing
flops
-> vague prompts
-> “do it all” asks
-> agents touching random files
net: split the work like chef (cursor) + sous-chef (codex) and you’ll ship faster
Top comments (0)