Skip to main content

System design answer

A coding agent that fixes small bugs

An agent with repository tools in a sandbox, tests as the main check, budgets, and a person reviewing every pull request.
The question“Design an agent that takes a bug ticket, finds the code, makes a fix, runs tests and opens a pull request.”

1Clarify0-5 min

Ask before you draw. These answers change the design.

  • What kinds of tickets: small bugs or features?
  • Which languages and how big is the codebase?
  • Can it run tests and install packages?
  • Who reviews and merges?

2Requirements and numbers5-8 min

It must

  • Read the ticket and search the code
  • Edit files and run tests in a sandbox
  • Open a pull request with an explanation

And it must be

  • Never pushes to the main branch
  • No secrets or network in the sandbox beyond the package mirror
  • Cost and time capped per ticket

Numbers to say out loud

Tickets
200 small bugs a week
Steps
20-60 tool calls, up to 15 minutes
Success target
About 40% merged without changes

3High-level design8-12 min

  1. Bug ticket → Agent loop
  2. Agent loop → Search and read code
  3. Agent loop → Edit files in a sandbox
  4. Agent loop → Run tests in a sandbox
  5. Run tests in a sandbox → fail → Agent loop
  6. Run tests in a sandbox → pass → Open a pull request
  7. Open a pull request → A person reviews and merges

4Deep dives12-26 min

Context

The agent searches the repository instead of loading it, keeps notes of what it has learned, and drops old tool output, so long runs stay within the context window.

Sandbox

A container per ticket with the repository, no secrets and network only to the package mirror; CPU and time limits; destroyed afterwards.

5Failures26-30 min

What can go wrongWhat stops it
Fix passes tests but is wrongHuman review; the agent adds a test that fails before the fix
Endless loopsStep and time caps
Dependency confusion or malwarePackage mirror only; no internet

6Evals and monitoring30-33 min

  • A set of 100 past bugs with known fixes: share solved, tests added
  • Review acceptance rate and reviewer time saved

7Trade-offs33-35 min

  • More autonomy saves more time but needs a stronger sandbox
  • Small tickets first; features later

Follow-ups to expect

  • How would you stop it from deleting tests to make them pass?
  • How do you measure value for the team?

Practise out loud in the mock interview

Today's AI turns are used

Lessons, roadmaps, videos, practice cards and the code playground stay free with your account, with no limit.