Hiring assessments built for a world where every candidate has a frontier model open in the next tab.
A take-home used to be the most honest signal in hiring: hand someone a hard, realistic task and read what they produce. That signal is gone. Fluent code, polished analysis, and a confident write-up are now free — a static assignment graded on its deliverable no longer measures the candidate. It measures their chatbot.
The reflex — ban AI — fails twice. It's unenforceable in a take-home. And it screens for a working condition that won't exist on the job: your hires will work with AI every day. The question isn't whether a candidate used it. It's whether they were more than a relay.
We design assessments where AI use is permitted, expected, and genuinely useful — and where the score still depends on the things a model doesn't supply on its own: deciding what to investigate, which information to trust, what to verify before committing, and what to ask when the materials run out.
Those are judgment calls, and they're the same calls your strongest people make every day. An assessment that isolates them measures something durable: how a candidate steers powerful tools through an unfamiliar problem — not how well they prompt.
Each assessment is a self-contained world built from scratch, with a measurable ground truth underneath it. Nothing about it appears in any training data, and performance is scored against reality — not against a grader's impression of the write-up.
The information needed to do well is discoverable in the materials but never handed over. Finding it, validating it, and acting on it is the test — work that doesn't happen unless the human directs it.
Before any candidate sees an assessment, we run the strongest available models on it, working alone — and tune until the gap between an unaided model and a person exercising real judgment is wide and measurable. As models improve, we re-run and re-tune. The instrument stays sharp instead of silently going stale.
The deliverable is scored against ground truth. The process is graded alongside it: what the candidate examined, what they questioned, what they verified, what they asked. You see how someone works, not just what they produced.
A scored, ranked result your team can act on without redoing the work. A short reasoning-trace read on each candidate — how they explored, what they caught, what they missed. And an assessment that's maintained, not shipped and forgotten: re-calibrated against each model generation, so the thing it measures this year is still the thing it measures next year.
Work tests run standalone, or as the assessment stage inside an embedded engagement.
For inquiries: vaishnav@cloutcareers.com