Notes from building agent-allowed, understanding-verified hiring — challenge design, the reasoning defense, and why we stopped asking whether a human typed every line.
5 min read
hunr doesn't detect or block AI use on its challenges, on purpose. The signal isn't whether a human typed every line — it's whether this person can ship, reason about, and own real work in your stack.
5 min read
The strongest hiring signal in the agent era is the gap between how polished a submission looks and how well its author can defend it. hunr's reasoning defense measures that gap directly, with questions generated from the candidate's own code.
6 min read
Greenfield plus well-specified is the agent sweet spot and a dead hiring signal. The durable design levers: messy realistic codebases, implicit-but-critical requirements, multi-step chains, and mutation-gated hidden tests.
5 min read
hunr v1 generated challenges with AI and graded them with deterministic tests, no AI in the execution loop. It proved that work samples beat resumes. Then the agent era exposed the assumption underneath it: that a passing artifact proves understanding.
4 min read
Cheating attempts on coding assessments roughly doubled to ~35% in 2025, vendors admit they can't reliably catch AI assistance, and the industry's leaders have already pivoted: allow AI, grade understanding.
5 min read
I built hunr after sitting through interviews and watching the same failure from both sides: resumes that promised more than the candidate could ship, and great engineers filtered out before anyone saw their work.