Working with Coding Agents
What changes when the assistant stops suggesting and starts acting.
The difference that matters
A chat window is a conversation. An agent in your project opens files, changes them, runs commands and reads the output.
The speed is real, and so is the review it quietly skips.
- Mistakes land in the repository, not in a paragraph you can ignore.
- Commands run: installs, deletions, rewritten history.
- Twenty minutes of agent work outpaces a day of reading.
Five minutes of setup
Most of the ways a first week goes badly are removed before the first prompt.
None of this is about prompting. It is about keeping mistakes cheap.
- Start from a clean tree, so the agent's work is its own diff.
- Work on a branch you are willing to abandon.
- Know the exact command that gets you back to now.
- Decide in advance what it may run, not mid-flow.
- Keep credentials outside the working tree.
The loop that works
The shape of a good session is the same whichever tool you use.
Small steps are not about the agent's ability to hold them. They are about yours to check the result.
- State one task with the finish line named.
- Ask for the plan before the edits.
- Let it work in a bounded step, not on an open mandate.
- Read the diff — every changed line, before anything else.
- Ask for proof: the output, not “should work now”.
- Commit small and often; commits are the checkpoints.
When to reset instead of retry
Go back to the last commit and say what actually needs to change. Agents do not get tired, so they will keep layering fixes on a broken approach for as long as you let them.