Field notes
Long-form arguments on AI cheating, interview design, and what it now takes to actually measure technical capability. Written for hiring managers, talent leaders, and engineers who are tired of the theatre.
You cannot stop a competent candidate from running an AI overlay during a take-home. The defensible move is to design assessments where AI use is either impossible or expected and instrumented. Here is the practical playbook.
Read article →A Capterra survey put the rate at 47%. The University of Manchester caught 150 students in a single November sweep. Here is what the data shows about how candidates are cheating with AI — and what most hiring teams still get wrong about it.
Read article →Interview Coder ships an undetectable Electron overlay that reads the question off your screen and pipes a solution back. Cluely is the more polished cousin. Most legacy proctoring stacks cannot see them. Here is what they actually do, and what does still work.
Read article →A paid trial week for a senior contractor costs you £4,000–8,000 before you know whether to keep them. AI has broken the take-home as a triage filter. Toptal charges a 30% markup to do the triage for you. There is a third path.
Read article →There are exactly two stable equilibria for a 2026 coding interview: questions that an AI cannot help with, or questions where AI use is required and instrumented. Everything in between is a slow-motion failure.
Read article →Identical constructs, unique questions. A CV-grounded interview keeps the rubric structured — so the scoring is fair across candidates — while the surface questions stay unique enough that leaked answer keys do not work.
Read article →Harvard and BCG ran 758 consultants through GPT-4. The ones who used it well outperformed the ones who used it badly by a wider margin than the average performance gap between consultants. No mainstream interview measures this skill.
Read article →Schmidt and Hunter’s 85-year meta-analysis is still the cleanest empirical answer to “what predicts job performance.” Structure wins. AI can add more structure than any human panel ever could. It can also amplify bias if you do not design for that explicitly. Both are true.
Read article →We do 20-minute intros where we listen first. No deck.
Book a call