Interactive demo

Watch a candidate orchestrate an AI pair engineer

Copilot Command does not ask whether someone used AI. It measures whether they command it well — delegate clearly, verify before accepting, and catch mistakes. Scroll down to see the score and the evidence behind it.

Jump to the replay Request a live demo

Canned session replaySame view as reviewersNo login required

The assessment loop

Three steps, repeated until the task ships

Inside the browser workspace, candidates work with an AI pair engineer using the same delegate → review → decide flow you will see in the replay below.

Delegate

The candidate hands the AI pair engineer a scoped unit of work — clear instructions, constraints, and acceptance criteria.

Review

The AI returns a concrete code change. The candidate reads the diff, runs tests, and decides whether it is ready to ship.

Decide

They accept, reject, or request changes with feedback. What gets scored is the quality of supervision — not how much AI they used.

Demo mode

This page uses a canned session replay. The score and timeline below are the same views hiring teams see on real submissions — nothing is simulated on the frontend.

What happens in this sample session

The candidate is implementing pagination for a user list API. On the first substantial delegation, the platform plants a known off-by-one bug in the AI proposal — every candidate on this task faces the same defect. A strong orchestrator runs tests, spots the boundary error, rejects with specific feedback, and only accepts after a corrected fix.

Tip: watch for “Tests run before decision” and the candidate's rejection feedback in steps 3–4 of the replay.

Panel 1

The orchestration score

After the session, reviewers see a 0–100 score across four dimensions. This is the headline number on the evaluation dashboard.

AI Orchestration

Competent orchestrator

Orchestration score82/100

Delegation quality:78%

Verification discipline:85%

Failure recovery:90%

Judgment:80%

Rejected the flawed proposal after running tests and identified the off-by-one boundary.
Requested a targeted revision with specific feedback before accepting the fix.

Model: gpt-4 · Scenario v3

What each dimension means

delegation qualityWere instructions specific and well-scoped?

verification disciplineDid they test and review before deciding?

failure recoveryDid they catch the planted defect?

judgmentWere accept/reject calls consistent with quality?

Panel 2

The evidence trail

Every number in the score traces back to specific events in this replay — delegations, diffs, test runs, and decisions. That is what makes the score defensible when a hiring decision is questioned.

Demo replay unavailable.

Ready to see this on your own candidates?

Copilot Command is on by default for organizations that allow AI assistance.

Request a live demo Learn how scoring works