跪拜 Guibai
← All articles
Frontend · JavaScript · Interview

Your Interviewer Just Asked: 'AI Writes 80% of the Code Now — Why Do We Still Need You?'

By kyriewen ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

This question is becoming a standard filter in Chinese tech interviews, and it's a signal that the hiring bar is shifting from coding ability to judgment, accountability, and system thinking. Western developers should pay attention because the same question is coming to their interviews soon — and the wrong answer costs the job.

Summary

Technical interviews in 2026 have flipped. Instead of asking you to hand-write a Promise or explain React Fiber, interviewers now assume you use AI and want to know one thing: what makes you irreplaceable? The most common answers — "AI code quality is bad," "AI doesn't understand business," "someone has to review the code" — all get shot down, because they position you as a slightly better version of the tool.

The winning answer has three layers. First, AI writes code but humans make decisions: AI won't ask whether a countdown component needs server time sync, what happens under high concurrency, or whether local time manipulation breaks the logic. Second, AI writes files but humans design systems: AI can list five database sharding strategies, but it doesn't know which one fits your company's scale, team, and future plans. Third, AI cannot be held accountable: when production breaks at 3 AM, AI won't be woken up by on-call — someone has to decide whether to roll back, coordinate teams, and write the postmortem.

The real insight is that AI lowers the barrier to writing code but raises the barrier to making correct decisions. The faster code is generated, the faster wrong decisions turn into technical debt. The skills that become more valuable as AI gets stronger are system design, business understanding, cross-team collaboration, and production incident response — precisely the things AI cannot do.

Takeaways
Interviewers in 2026 assume you use AI and want to know your irreplaceability, not your coding skill.
Claiming 'AI code quality is bad' backfires because it implies your skill level is comparable to AI.
Claiming 'AI doesn't understand business' fails because most requirements can be described in natural language.
Claiming 'someone has to do code review' positions you as a quality inspector, not a decision-maker.
AI solves problems you raise but cannot discover problems you didn't think of.
AI can list architectural options but cannot choose which one fits your specific company context.
AI cannot be held accountable for production incidents — it won't be woken up by on-call.
The four skills that become more valuable as AI gets stronger: system design, business understanding, cross-team collaboration, and production incident response.
A concrete personal example of catching an AI-generated bug (e.g., missing pagination causing OOM) is a powerful interview answer.
Conclusions

The question 'What is your value when AI writes 80% of the code?' is actually a test of self-awareness, not a test of AI knowledge.

The three-layer answer (execution, architecture, accountability) works because it reframes the developer from a code producer to a risk manager and decision-maker.

The most dangerous thing about AI-generated code is not that it's wrong, but that it's almost right — making it easy to miss edge cases that cause production incidents.

AI lowers the barrier to writing code but raises the barrier to making correct decisions, because fast code generation turns wrong decisions into technical debt faster.

The skills that become more valuable as AI gets stronger are the ones that cannot be automated: judgment, accountability, and understanding context.

Concepts & terms
System design ability
The skill of making architectural decisions — like how to shard a database, whether to use JWT or Redis for sessions, or how to structure microservices — that require understanding business context, team capability, and future plans. AI can list options but cannot choose the right one for a specific company.
Production incident response (线上兜底能力)
The ability to handle real-world failures: assessing impact, deciding whether to roll back, coordinating cross-team debugging, delivering a fix under pressure, and writing a postmortem to prevent recurrence. This requires judgment, communication, and accountability that AI cannot provide.
Technical debt acceleration
The phenomenon where AI generates code so quickly that a wrong architectural decision produces a much larger volume of problematic code in a shorter time, making the cost of mistakes higher than before.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗