In real interview logs circulating on developer communities Niuke, Juejin, and SegmentFault, traditional coding questions are being swapped out: "implement quicksort by hand" and "explain the event loop" are giving way to prompts like "AI-generated code ships and triggers a P0-level (P0 = highest severity) incident—how do you split liability?" "A reviewer marked 'verified' but missed a bug—who's on the hook?" "Do you treat AI as a typewriter or as a junior intern?" The reason for the shift is straightforward. Once "shipping code" stops being scarce and every resume lists a few AI-built projects, interviewers face a new problem: "Everyone looks stronger on paper, but the difficulty of actually filtering out signal has gone up." AI has leveled implementation ability; judgment, verification, and accountability are three things it cannot replace. What we should care about: enterprise AI is moving from the "can we use it" stage to the "who's responsible when it breaks" stage. Coding is just the first profession to hit this wall.
Industry View
The supportive camp says this signals enterprise AI governance is starting to land—a mark of maturity. The "typewriter vs. junior intern" framework at least offers a layered approach: the submitter carries baseline responsibility, the reviewer verifies, and the process retains an evidence chain (complete, traceable decision records). The fact that interview questions can pose this problem means organizations are already thinking about it. But the objections and risks are equally clear. First, these 12 questions themselves prove that even the most AI-savvy tech companies haven't figured out the answer, let alone other enterprises. Liability allocation is still one standard answer short of the casual summaries floating around. Second, "typewriter vs. intern" is a metaphor, not a policy. Apply it to non-code AI applications—contracts drafted by AI, financial analyses run by AI—and no one can articulate where the breach point sits or how to slice responsibility. Third, this rise in interview screening difficulty could end up slowing down enterprise AI adoption: too few teams can handle the accountability question, so most companies will prefer to wait and see.
Impact on Regular People
For enterprise IT: AI governance will land on the compliance checklist. Practices like "specialized audits of AI-generated content" and "evidence chain logging" will migrate from technical documentation into management policy. For individual careers: even if you don't write code, anyone using AI to produce slide decks, contracts, or financial analyses faces the same question—when AI breaks something, can you personally bear the responsibility? The underlying logic is identical. For the consumer market: AI tools' disclaimers will increasingly resemble software license agreements. Ordinary users will need to start reading the fine print and understand what AI cannot underwrite on their behalf.
牛客、掘金、SegmentFault 这几个程序员社区上能查到的真实面经里，传统考题正在被替换：「手写快排」「说一下事件循环」变成「AI 生成的代码上线后出了 P0 级（P0 = 最高严重等级）事故，责任怎么分？」「review（同事复核）的人标了'已验证'，但他验证漏了呢？」「你把 AI 当打字机还是当初级实习生？」变化的原因不复杂。当「写出来」不再稀缺，简历上人人都有几个 AI 做的项目，面试官面对的新问题是「大家看着都变强了，但真正能筛出来的难度反而上升了」。拉平的只是实现能力，判断、验证、担责这三件事 AI 替代不了。值得我们关心的是：AI 在企业里正从「能不能用」阶段进入「出事谁负责」阶段。代码只是最先撞上这堵墙的工种。
行业怎么看
支持的声音是：这说明企业 AI 治理开始落地，是成熟标志。「打字机 vs 初级实习生」这个框架，至少给出了分层思路——提交者担底，review 者验证，流程留证据链（完整可查的决策痕迹）。面试题能问出这个问题，意味着组织里已经在想这件事。但反对意见和风险同样明显。第一，这 12 道题本身就证明，连最懂 AI 的科技公司都没想清楚答案，遑论其他企业。责任分担比坊间的总结还差一个标准答案。第二，「打字机 vs 实习生」是个比喻，不是制度，套到非代码的 AI 应用——AI 写的合同、做的财务分析——失守点在哪、责任怎么切，没人说得清。第三，这波面试筛选难度上升，可能反过来拖慢企业用 AI 的节奏：能压住责任问题的团队太少，多数企业更愿意观望。
对普通人的影响
对企业 IT：AI 治理会进合规清单，「AI 生成内容专项检查」「证据链留痕」这些动作会从技术文档爬到管理章程。对个人职场：哪怕你不写代码，用 AI 做 PPT、合同、财报分析的人也要面对同一道题——AI 出的事，你担不担得起。逻辑是一样的。对消费市场：AI 工具的免责说明会越来越像软件许可协议。普通用户要开始习惯读条款，知道哪些事 AI 不能替你兜底。
