What this is
ShowMeBug, Cursor, Claude Code — these AI tool names are now appearing on the evaluation sheets of top-tier programmer interviews in China. Candidates are dropped in front of a broken frontend project, open an AI tool, and fix bugs on the spot. Interviewers evaluate not "can you code" but "can you make AI produce working code."
A Juejin piece summed up three moments when candidates "got cut off": immediately asking AI to rewrite the whole file, accepting AI output without strict review, and submitting vague prompts that fail to clarify requirements. Behind those failures sit three capabilities — problem localization, critical verification, and clear expression. That is what the interview is actually testing.
Industry view
The hiring community broadly approves: the old algorithm-quiz filter weeded out plenty of production-grade engineers. Real tasks plus AI tools better reveal who can actually ship.
But dissent exists. One view argues this is companies quietly squeezing headcount — once "knows AI" becomes the bar, one senior engineer plus AI can replace two or three junior roles, narrowing the entry path for new grads and career-changers. A more concrete concern: AI-fixed code runs in a demo environment, but production edge cases are far more complex. Treating "knows AI" as equivalent to "can ship" may plant the seeds of production incidents.
Impact on regular people
For enterprise IT/HR: technical hiring criteria are being rewritten. Items like "familiar with Git/CI" are losing weight; "can localize problems, can review AI output, can write requirements clearly" is becoming the new baseline.
For individual careers: the yardstick shifts from "how much knowledge you have memorized" to "how much quality output you can drive from AI." Pure-recall roles are devaluing fast; the soft-hard hybrid skill of "decomposing vague requirements and judging whether AI output is correct" is rising in value.
For the consumer market: market-education costs for AI coding tools like Cursor, Claude Code, and Trae are being lowered by free endorsement from big-tech interviews. But users will grow pickier — the gap between tools that "actually solve engineering problems" and tools that "just generate code that looks right" will widen quickly.