Stop interviewing developers. Start auditing their repos
The whiteboard interview tests for a skill nobody uses. The repo audit tests for the skill you actually pay for. Here is how to run one
The technical interview tests how someone inverts a linked list under stress. The job tests how they control a function with Cursor while the model hallucinates an import. These are not the same skills and they correlate roughly with zero
What to audit instead
Choose their three most recent bundled PRs. Read the difference. Read the description. Read the review comments. You're looking for four things
Application discipline, does the PR do one thing or seven?
Naming... can you understand the change without the description?
Test coverage... did they write a test that would have found the bug, or one that only exercises the happy path?
Assessment signal, did senior engineers push back and did the candidate argue well or did he succumb?
What the whiteboard misses
The whiteboard tests a state in which no one is working. No IDE, no Cursor, no Claude Code, no Stack Overflow, no time to think. It is a test of stress tolerance disguised as a test of skill. 92% of US developers use AI tools at work and you test them on the 0% of work that excludes it
Then you wonder why 83.9% of IT projects fail or overrun. They passed a test for a job that doesn't exist
The AI-era addition
Ask them to review a recent Cursor or Claude Code session with you. What prompts did they discard? What did they rewrite by hand? A true vibe coder rejects half of what the model proposes. A LARPer accepts everything
You don't need a loop of six rounds. Thirty minutes of repo auditing is always better than eight hours of interviewing. We know that because we based our whole vetting process on that
The objection from HR
"We can't control private repos. You can. Ask the candidate to go through one with you on their machine, screen shared. Read it live. If they refuse, that's a signal. Real senders like to show their work. Those who hide it are hiding something
Add the 30 minutes back
81 days to hire, $112K burned per seat, 83.9% project failure rate. Repo audits reduced all three. Faster decisions, fewer false positives, better hires. The math is embarrassing to anyone still working with six-round loops