Agentic coding / 2 MIN READ
Better context. Better changes.
Turn a vague request into a focused task with clear boundaries and a definition of done.
A useful task brief lets another developer understand the work without guessing. The same habit helps when that collaborator is a coding agent.
Start with the problem
Describe what happens now and what should happen instead. Include a way to reproduce a bug, or a concrete example of the desired behavior.
“Fix the mobile menu” leaves many decisions open. “At 390 pixels wide, the menu covers the search field and cannot be dismissed with Escape” gives the agent something observable to investigate.
Provide context that changes the decision
Share the relevant files, the project’s conventions, and the constraints the solution must respect. Useful context might include an existing component to reuse, a browser requirement, or a route that must keep working.
Avoid filling the brief with unrelated files. Keep credentials and private production data out of examples; use representative sample values.
Use a repeatable template
Problem:
What currently happens, and who is affected?
Expected behavior:
What should the user see or be able to do?
Relevant context:
Files, conventions, examples, and commands worth inspecting.
Constraints:
What must remain compatible? What is outside this task?
Acceptance checks:
Concrete cases that demonstrate the change works.
Let the codebase inform the plan
Ask the agent to inspect the existing implementation before editing. It may find a reusable pattern or discover that the behavior comes from a different part of the system than you expected.
If new information changes the scope, update the brief. For a larger task, divide work into independently verifiable steps: data loading, rendering, then interaction.
Make completion observable
For a search feature, acceptance checks might be: a known query returns a result, a nonsense query displays the empty state, clearing the query restores results, and the page remains usable with a keyboard.
Ask for a final explanation of what changed, how it was checked, and what remains uncertain. Those three points are useful whether the work took five minutes or several hours.