The Last Technical Interview

Source: Steve Yegge Date: 2026-05-29 URL: https://steve-yegge.medium.com/the-last-technical-interview-bc13ddcf4564?source=rss-c1ec701babb7------2

Summary

Steve Yegge’s essay argues that the technical interview — unchanged for roughly 50 years — produces assessments barely better than random: interviewers show minimal agreement on the same candidate, and Google’s own internal calibration exercise revealed that evaluators would reject approximately two-thirds of their own colleagues. The essay identifies internships, co-ops, and contract-to-hire as the only evaluation formats with a demonstrated track record, and proposes a “campfire” model where candidates do paid real work with portable, company-signed credentials as the output. Yegge notes that some San Francisco companies are already experimenting with this informally.

Implications

The signal feeds the coding-agent landscape and developer labor market threads directly. If the campfire model gains traction, the leverage point shifts from interview performance to demonstrated agentic output — work artifacts that are credentialed and portable. This is structurally aligned with where coding agents are heading: agents already produce signed, reviewable commit histories and PR records, which are closer to Yegge’s “real work” standard than any whiteboard session. Implications for the tooling layer:

• The move from simulated-work interviews to real-work assessment creates demand for tooling that makes agentic output legible and attributable — audit trails, contribution graphs, and agent-assisted work logs become resume artifacts.
• Coding agents lower the cost of producing polished work samples, which could paradoxically accelerate the collapse of traditional interview signaling (if everyone’s output looks the same, proof-of-process becomes the differentiator).
• Teams building agent-assisted hiring or onboarding tools have a clearer thesis here: the evaluation format is changing before the tooling has caught up.