I have been using Codex CLI as my daily driver for AI-assisted coding for a few months now, and I keep coming back to a handful of patterns that made my work faster and less error-prone. This post is a short, opinionated tour of the workflow that finally clicked for me.
Why another AI coding tool?
I bounced off several chat-only assistants because they forget context the moment I switch files. Codex CLI runs in the same terminal where my project lives, so it can actually read what I am looking at, run tests, and propose diffs I can review line by line.
The three commands I actually use
codex "explain this function in plain English"-
codex --editto let the model propose and apply small refactors -
codex --planwhen I want a written plan before any code changes
Lessons from the trenches
- Always commit before letting it edit. It is rare, but it does occasionally nuke something.
- Treat the first response as a draft, not an answer. I almost always ask for a second pass.
- Keep the change scope tiny. The smaller the diff, the better the review.
What is next
I am still figuring out how to use it well for non-code tasks like writing tests, reviewing PRs, and triaging issues. If you have a workflow that works for you, I would love to hear it in the comments.
Top comments (0)