Interviews Request access

Hire engineers who can steer AI.

Candidates fix a live broken system with AI. You see the proof.

Request access See a sample report
The task

Checkout fails at random when traffic is high. Find out why and fix it.

Code
} catch (e) {
  if (e instanceof PaymentTimeout) {
    return { status: 'pending' };
  }
AI assistant
CandidateCheck the connection pool before touching code.
AIAll 10 connections are in use, none free. Looking for one that is never given back.
Live logs
warn payment timeout, queued
error timeout exceeded when trying to connect
error checkout failed, 500

An interview environment built for the AI era

Interviews is a hosted workspace where each candidate gets a real broken system, a code editor, a terminal and an AI agent.

You choose the problems from your own stack. We run the session and hand you a report on how they investigated, directed the AI and proved the fix.

Your candidates use AI every day. Your interview pretends they don’t.

Most interviews test

  • Recalling an algorithm without help
  • Writing code from a blank file

The job now needs

  • Finding the cause in messy logs
  • Directing an AI agent and catching its mistakes
  • Proving the fix works

A hiring signal you can defend in the debrief

See the thinking

Watch how they frame the problem, question the AI and prove the fix, not just the final answer.

Trust the score

Hidden checks decide what is really fixed. Every rating links to evidence you can read.

Give engineers their hours back

No more live puzzle sessions run by your team. The report is ready when the session ends.

Three steps from your stack to a decision

  1. We build the problem from your stack

    Tell us your languages and services. We build realistic problems from your stack. You pick the final few.

  2. The candidate works in a browser room

    Editor, terminal, AI assistant, logs and tests. Nothing to install. No video call.

  3. You get a report, not a feeling

    Scores for how they investigated, directed the AI and verified, each tied to what happened.

Every score points at something they did

  1. 12:04Asked the AI to reproduce the failure before changing anything.
  2. 27:31Rejected an AI patch that never gave the connection back.
  3. 41:55Wrote a test for one bug and missed the other.
Senior backend engineer
Sample with illustrative data
Lean hire
  • Checked logs and metrics before touching code4/4
  • Gave the AI specific, scoped requests3/4
  • Caught a wrong AI patch3/4
  • Verified the fix with a load test2/4
  • PassNo failures under load
  • PassNo leaked connections
  • FailCustomers still charged twice

Your stack, inside the room

The room mirrors the tools your engineers really use, so candidates debug the way they would on the job. Solid tiles are live today. Dashed tiles are built for your pilot.

In the room

The tools a candidate works in: editor, notebooks, terminal and AI agent.

  • VS Code
  • opencode
  • Claude Code
  • JupyterLab

Your systems

What the broken system runs on and reports to, so logs, dashboards and data look familiar.

  • Postgres
  • Datadog
  • Grafana
  • Kubernetes
  • Snowflake

Your workflow

Where scenarios start and where the finished report is delivered.

  • GitHub
  • GitLab
  • Slack
  • Greenhouse
  • Lever
  • Ashby

Not on the list? We build it into your pilot. Logos belong to their owners and imply no endorsement.

Two rooms are ready. More come from your stack.

Senior and staff software engineer

The problem
Checkout fails at random under heavy traffic.
A strong candidate
Finds the leaked connection, then spots the double charge.

Senior and staff machine learning engineer

The problem
A ranking model got worse after a data pipeline change.
A strong candidate
Finds where training and live data disagree, then adds monitoring.

Data science, AI engineering and product rooms are next.

Fair to candidates. Reliable for you.

Same room for everyone
Identical start, one time limit, one AI budget. Candidates know what is recorded.
Problems with a known answer
We solve each one ourselves, with and without AI. A hidden grader checks results.
Contained and your choice of model
Each session runs in its own sandbox, on models your team approves.

See it on your own stack

We onboard teams by hand.