For those of you using AI coding agents for work, what's your typical workflow? Do you mostly stay inside the IDE, or are you primarily using the CLI or desktop tools like Cursor, Codex, and/or Claude Code?
At my company, we haven’t formalized anything yet. Our repos don’t have any standard files like AGENTS.md, skill definitions, etc. Everyone is basically doing their own thing.. if they’re using AI at all. We still have several people who refuse to touch it.
I’ve moved almost completely out of the IDE and now use the CLI almost exclusively. I set up a “sidecar” directory alongside the project where I keep all the configuration files, point it at the source code, and define plans, tasks, etc. Session continuity comes from files on disk rather than the agent’s conversation history. I'll verify what's written within the IDE, approve or issue revision requests.
This is the system I’ve come up with on my own. Curious to hear how others are approaching it.