AI FOR YOUR WORK · DEVELOPERS

Make AI your superpower: developers

Move from a reproducible issue to a reviewable fix.

Your first project: A small bug fix with evidence

A focused code change, a regression check and a concise explanation for the reviewer.

You bring the expertise: what matters to your audience, how your work is judged and what a useful result looks like. Give AI the organizing, drafting and exploration work that helps you move forward. Keep the decision in your hands.

Gather these inputs

A reproducible issue, expected and actual behavior, the relevant files, project conventions and the commands used to verify a change.

Choose one project that fits a focused working session. Save the original inputs and your final version together so you can see what changed. If you already pay for an assistant, begin there before adding another subscription.

1. Reproduce before editing

Give the agent a concrete input and the incorrect output. Ask it to trace the relevant code and explain the likely cause. If it cannot reproduce the failure, resolve the missing setup before accepting a speculative fix.

2. Define the smallest change

Agree which behavior must change and which interfaces must stay compatible. Ask the agent to name the files it expects to touch and why. A small patch is easier to inspect, debug and roll back.

3. Build a useful check

Add a test that fails on the original bug and passes after the fix. Cover an adjacent edge case where the same assumption could fail. Ask for the command and actual output, rather than accepting “tests pass” without evidence.

4. Review the diff

Read every changed file, including dependency and configuration changes. Check the behavior in the running interface when the task affects users. Keep repository review and deployment practices in place.

Try it with this practice brief

Practice inputA list filter treats the string "0" as an empty value, hiding a valid zero-price item. Expected: an empty string clears the filter; the string "0" searches for zero.

What a useful result preserves

A regression check covers both inputs separately. The patch fixes the empty-value check without converting every value through a truthiness shortcut.

Run the brief once, then deliberately remove an important detail. Check whether the assistant asks for it, leaves it unknown or invents an answer. Revise your instructions around the mistake you actually observe.

Copy your starting prompt

Investigate this reproducible bug in the supplied repository context. First explain the cause and propose a minimal change. Implement it using project conventions. Add a regression check that fails before the fix and covers the adjacent edge case. Report changed files, commands actually run, results and remaining uncertainty. Do not deploy.

MY CONTEXT AND SOURCE MATERIAL:
[Paste permitted inputs]

OUTPUT REQUIREMENTS:
[Format, audience and constraints]

Keep the process that earns its place

Track time to a reviewed fix, number of review corrections and whether the regression check catches the original failure.

Use the workflow on three representative examples. Keep a short correction log: what went wrong, what you changed and whether the next attempt improved. Save the best prompt with a good input/output example. That becomes a reusable working method for your next project.

When the steps become predictable, use the task automation guide. When the job needs an assistant to choose tools, follow the first-agent tutorial. Compare tools on the same brief in the Tool Trial Lab.

Common questions

Do I need a new subscription?

Start with an approved tool you already have. Add another only when the task exposes a specific missing capability. Check current plan limits and usage charges.

How do I know this is helping?

Count the time spent reviewing and correcting as well as drafting. Keep the workflow when it improves the finished result or makes the total process easier.

Explore the tools

GitHub MCP ↗Context7 ↗

Sources and review notes

An original practice workflow from AI Superpower. The linked reference provides product or workflow background; the example is an exercise you can adapt.

Keep learning

← All journal articles