The problem
A team's second-round interview is a timed algorithm puzzle. Strong candidates who now work through a coding agent every day find it disconnected from the job; the puzzle also says nothing about how they use AI, which is most of how the team writes code.
The team wants a round that looks like a real day: an unfamiliar repository, a bug report, a test suite, and the tools they would actually use.
The approach
- 01
Pick an issue that matches the role
The hiring manager browses the issue bank and pins one specific issue to the link, so every candidate for the opening works on the same problem and results compare like for like.
- 02
Personal links, so nothing is shared
Each shortlisted candidate gets a personal link bound to their email and usable once. The round runs in an isolated browser VS Code with an AI agent, so nothing has to be installed.
- 03
Debrief from the timeline
The follow-up conversation uses the report's timeline and prompt log: why did you read that file first, what made you rerun the tests there, why did you reject the agent's first patch.
What the report shows
The signals a team would read in this scenario, all recorded in every round. See them in the sample report.
- Exploration
- Which files were read before the first edit, and whether the candidate found the cause or patched the symptom.
- Recovery
- What happened after a failing test run or a wrong turn by the agent: a loop of retries or a change of approach.
- Speed
- Pace on work that landed, gated on the outcome, so fast-and-wrong never looks good.
- Test runs
- How often the candidate ran tests and when, visible on the timeline.
The takeaway
The round measures the skill the team hires for, and the debrief becomes a conversation about real decisions in a real codebase rather than a recital of a memorized solution.