FDE Times: AI-allowed engineering interviews now grade how you doubt and verify agent-written code

Anthropic’s engineering blog notes that each new Claude model forces a redesign of its hiring test because the model matches strong candidates; the company’s performance-engineering take-home still allows AI tools, as on the job. Meta and Google run AI-enabled coding rounds, and interview guides say some OpenAI Forward Deployed Engineer rounds — including Codex screens — do the same.

Across those loops the graded skill is less “can you write the code” and more “do you judge it.” Meta’s framing of critical verification, per techinterview.org and LeadDev, watches whether candidates read AI output before running it and grow more suspicious when an answer looks neat. OpenAI FDE guides emphasise screen-sharing, reasoning aloud, and stating the time and memory costs of whatever the tool produces.

That shift matters especially for FDEs: agent-written code that looks right but fails on messy customer data has production consequences, not just a lost interview point. The piece recommends rehearsing five steps before trusting generated code — read every line, say what it should do, hunt for breaking inputs, state complexity costs, then keep, fix or discard based on tests — and rewriting CV bullets to show a specific bug caught before production.