How to prompt an AI coding agent for better code

Give context, not just commands

The agent does better when it knows the why behind the what. "Add a save button" is thinner than "add a save button so users do not lose their work if they refresh." Context helps the agent make the small decisions the way you would.

Be specific about the outcome

Describe the result you want in concrete terms: what appears on screen, what happens when someone clicks, what the data looks like. Specific outcomes are easy for the agent to hit and easy for you to verify.

Iterate instead of over-specifying

You do not need a perfect prompt up front. Ask for a first version, look at it running, and refine. A short back-and-forth usually beats one enormous instruction, because you are steering with real feedback.

Describe problems precisely

When something is wrong, tell the agent exactly what you see and what you expected instead. "The list shows nothing after I add an item, but it should show the new item at the top" is far more useful than "it does not work."

Ask it to explain and check its work

You can ask the agent to explain its approach or to double-check for issues before you accept a change. Used well, that habit catches mistakes early and teaches you at the same time.

Frequently asked questions

Explore more