A thousand applicants a role — and none of the usual screens tell you who can actually do the job now. Hireproof measures it directly: can they direct an AI agent, catch it when it's confidently wrong, and stand behind what ships? On a real task, against the clock.
Writing code stopped being the bottleneck. Anyone, with any tool, can produce something that only looks right in seconds — the hard part is everything around it.
That's the real job now: knowing what to build from an ambiguous ask, catching what the AI gets wrong, and standing behind every line that ships. A candidate who can only generate code isn't faster than the rest — they're the new liability.
Real work starts from a vague ask and a capable agent — not a clean ticket. The job is turning that into the right thing to build, and pointing the AI at it.
How we prove itWe hand them a deliberately underspecified brief and a live agent, then watch: do they interrogate it, or ship the first thing it proposes?
AI writes code that looks right and often isn't. Catching what's confidently wrong before it ships is what separates an engineer from a code generator.
How we prove itOur agent plants subtle, realistic bugs on purpose. Your report shows exactly which they caught and which slipped through — measured, not guessed.
You're accountable for every line that ships — whether you wrote it or the AI did — and you have to know why it's right.
How we prove itThey're graded on the final code as their own. Then, in Part 2, a neutral examiner asks them to defend the decisions behind it.
The candidate gets a deliberately underspecified spec — and has to ask the right questions before they build.
They direct an AI agent that builds alongside them — and quietly plants subtle, realistic mistakes as it goes.
They review, correct, and ship in a full workspace — editor, live server, database — under a 60-minute clock.
Optionally, a neutral AI examiner reviews the frozen work and asks them to defend the decisions behind it.
You get an objective read on what they caught versus shipped, plus how they defended it and how they worked.
When the build clock stops, an optional second phase begins. A neutral AI examiner reads the frozen work and generates questions about the decisions the candidate made — probing what they missed and asking them to defend their calls. It's there to surface not just how they built it, but why.
You switched the pagination divisor to Math.ceil. What was wrong with the original — and how would you have caught it if the tests hadn't?
Integer division dropped the last partial page. I'd catch it with a total that isn't a clean multiple of the page size — the count comes back one short.
The retry handler still isn't idempotent. Was leaving that a deliberate call, or did it get past you?
Illustrative exchange — not a real candidate.
Two objective pillars, kept separate. Supervision is measured against mistakes we planted on purpose. Craft is a read on how they build. And when Part 2 is on, their defense lands alongside — their own words on why they made the calls they did. The AI elicits; you judge. No black-box score.
Defended the pagination fix in their own words. Couldn't justify the missing idempotency check — flagged for the panel.
Illustrative example — not a real candidate.
We'll walk you through an actual report and replay, and set you up to run Hireproof on your own pipeline.