Do not hire for 'anti-laziness' as a vibe.

Video caption

Do not hire for 'anti-laziness' as a vibe.

Avoid a hidden junior bug as the test.

Reviewers need the same anchors.

My rule: The action should alter a real decision.

#EricFieldNotes

Full written post / accessibility read

I want engineers who continue after a plausible AI result, but I would not score a personality label. In a short structured interview, I would ask what contract the generated tests miss, what independent result would check it, what deliberate counterexample should fail and how the release decision changes with a new fact.

A fictional event producer has a clean generated patch and local tests. The unresolved question is whether lagging consumers and queued old events preserve the business result. The candidate must trace actual data flow and find who owns the compatibility window. This tests engineering reasoning and initiative together.

Give candidates the same short artifact, open questions and late fact. Ask two trained reviewers to score a sample of system reasoning, independent-test choice, revision and handoff; compare disagreements. Probe technical fundamentals with a code excerpt or follow-up. A candidate who speaks eloquently but cannot explain replay should not pass.

My rule: measure the candidate's next discriminating move after the agent's first green report, not their confidence or number of prompts. Do this because when agents produce convincing work quickly, the valuable human behavior is discovering what remains unproved and deciding whether to ship.

#EricFieldNotes

Four-beat scene transcript

1. Do not hire for 'anti-laziness' as a vibe.

I want engineers who continue after a plausible AI result, but I would not score a personality label. In a short structured interview, I would ask what contract the generated tests miss, what independent result would check it, what deliberate counterexample should fail and how the release decision changes with a new fact.

Visual: Make the verification behavior observable.

2. Use a substantive boundary.

A fictional event producer has a clean generated patch and local tests. The unresolved question is whether lagging consumers and queued old events preserve the business result. The candidate must trace actual data flow and find who owns the compatibility window. This tests engineering reasoning and initiative together.

Visual: Avoid a hidden junior bug as the test.

3. Calibrate against role work.

Give candidates the same short artifact, open questions and late fact. Ask two trained reviewers to score a sample of system reasoning, independent-test choice, revision and handoff; compare disagreements. Probe technical fundamentals with a code excerpt or follow-up. A candidate who speaks eloquently but cannot explain replay should not pass.

Visual: Reviewers need the same anchors.

4. Seek falsifying evidence, then own the choice.

My rule: measure the candidate's next discriminating move after the agent's first green report, not their confidence or number of prompts. Do this because when agents produce convincing work quickly, the valuable human behavior is discovering what remains unproved and deciding whether to ship.

Visual: The action should alter a real decision.

Research and claim limits

• U.S. OPM: Work Samples and Simulations (S151)
• U.S. OPM: Designing an Assessment Strategy (S156)
• Google Research, Towards AI as a Collaborative Partner (S155)

The examples identified as illustrative or simulated are design probes, not reported incidents. Vendor specifications do not establish workload performance.