Ideas on technical assessment, AI in hiring, and building better evaluation systems.
In July 2026, US employers still listed 7.3 million job openings, yet the hiring rate was just 3.2%, near its lowest since 2020. Applications per recruiter are up 412% since 2022. This is what's breaking between candidates and companies, and what replaces the shortcuts filling the gap.
A company that sells cheating detection published its own numbers: 38.5% of candidates were flagged, and more than 61% of those flagged passed anyway. Meanwhile detectors falsely flag non-native English speakers at several times the rate of native ones, and sixty universities have switched them off. Here is what the evidence says, what we sell and why, and what answers the question detection was always failing to answer.
Google measured its own interviews and found zero relationship with job performance. The algorithm interview has never been isolated and validated in the peer-reviewed record, not once. Here is what the research hierarchy actually shows, why AI finished off whatever was left, and what work-sample assessment replaces it with.
Google's own research found essentially zero correlation between algorithm interview scores and on-the-job performance. Work-sample tests outperform every other hiring method. So why is the industry still clinging to LeetCode? We dig into the data, the hidden costs, and the alternative that actually predicts engineering success.
89% of engineers use AI tools daily. Assessment scores have inflated 34% while job performance stayed flat. Banning AI is unenforceable. Ignoring it is negligent. We make the case for the harder third option: designing assessments where AI usage itself becomes a signal of engineering judgment.
Every platform claims their assessments are realistic. Almost none deliver. We introduce a five-dimension fidelity framework for evaluating assessment quality, environment, codebase, task, tools, and evaluation, and show why fidelity matters more than difficulty.