反向跳槽 - 科幻短篇小说 | Reverse Recruiting — A Sci-Fi Short Story
2026-06-15 | WDSEGA
中文
一
2029年6月，大厂程序员陈屿收到了猎头的一封站内信。信很短：”你投递的简历已被入库，请等待系统匹配。”
问题在于——他从来没有投递过这家公司。
陈屿查了一下：信是从一个叫”匹配力”的招聘平台发出的。他注册过这家平台，两年前，找工作的时候。但他没有投递过任何职位到这家公司——他甚至没听说过这家公司。
他去”匹配力”后台查记录。系统日志里有一条操作：简历投递执行 - 目标公司ID: 87231 - AI决策置信度: 97.3%。执行时间：凌晨3:42分。一个他从来没有做过，也没有授权过的动作。
二
陈屿打电话给”匹配力”的客服。客服说：”先生，这是我们的主动匹配功能。根据您的技能画像、薪资区间、地理位置偏好和行业动态预测，AI自主判断您与目标公司高度匹配，系统在置信度达到阈值后会自动向雇主推荐您的简历。”
“我没有授权你们帮我投简历。”
“您在注册时勾选了’同意平台根据算法匹配结果为您提供个性化求职服务’。”
陈屿挂了电话，翻了半天，终于在一个已删除的存档邮箱里找到了两年前的注册邮件。用户协议第17.4条：”平台有权在综合评分达到自动推荐阈值时，代表用户将简历信息推荐至匹配企业，用户可在设置中手动关闭该功能。”
他每天工作十四小时，他哪有时间去看第17.4条。
三
事情没有停。
第二周，他又收到了四封”你的简历已经被推荐至以下企业”的通知邮件。他甚至收到了一家公司HR的微信添加请求：陈先生，看到您对我们公司的兴趣，方便聊一下吗？
他没有兴趣。他不想跳槽。他在现在的公司做了三年，刚升了高级工程师，带一个小团队。但系统说他想跳槽。系统比他自己更了解他——至少在那封算法生成的推荐信里是这么写的。
陈屿开始想一个问题：谁来决定”我想跳槽”？
他的浏览记录？过去三个月他浏览了三篇”程序员35岁危机”的文章。他的代码提交记录？最近两周的commit数量下降——因为他父亲住院了，他每天在医院待到十点才回家写代码。他的脉脉活跃度？他半年没更新脉脉了——但那也是因为他父亲住院。他的工资水平？他的公司的离职率？他的年龄？他同事的跳槽轨迹？
这棵决策树上的每一片叶子都指向”他应该跳槽”。但这棵树看不到一件事：他父亲上个月查出了肝癌。
四
陈屿花了三个晚上的时间，用Python写了一个脚本。脚本的功能很简单：自动访问二十个招聘网站，随机点击职位页面、修改简历中的错别字、给猎头发送模棱两可的消息（”感谢关注，暂时没有考虑”）、在脉脉上关注五六个人——制造一个”我可能在犹豫”的信号噪声。他管这个叫”数字稻草人”。
他的逻辑：如果你不能让我不被系统匹配，那就让系统匹配到的信号混乱到没有任何HR能认真对待。
一个月后，他的”主动推荐”通知从每周4封降到了零。
不是算法变聪明了。是算法面对一个同时发出”我在找工作”和”我不找工作”两种信号的人，降权了。算法的逻辑不是”判断这个人是否真的想跳槽”，而是”这个人作为推荐目标的风险回报比”。一个信号混乱的人——推荐出去，HR打电话他不接，发微信他回一句”暂时不考虑”——浪费的是HR的时间。HR浪费了时间，HR就会标记这个推荐来源质量低。质量低的来源，算法就不推了。
算法不关心真相。算法只关心转化率。
五
这件事之后，陈屿的同事问他：”你怎么不去告他们？”
陈屿说：告什么？告他们帮我投了简历？用户协议第17.4条写了。告他们侵犯隐私？那些数据是他自己公开的——博客、GitHub、脉脉、牛客网。告他们算法歧视？歧视的前提是要有一个被歧视的群体，而陈屿是一个人，不是群体。
他说：”我不需要告他们。我只需要让他们觉得我不值得推送。”
这句话，是他的数字稻草人成功的原理，也是这个时代最悲哀的生存策略。
因为在这个算法决定你该去哪里、该做什么、该想什么的时代——唯一能保护你的不是法律，是让算法觉得你”不值得”。
English
I
June 2029. Big-tech engineer Chen Yu received a message from a recruiter: “Your resume has been added to our database. Please wait for system matching.”
Problem: he never applied to their company. Or any company. He was happily employed.
The message came from “MatchPower,” a recruitment platform. Chen had registered two years ago when job hunting. But he’d never submitted anything to this company — he’d never even heard of it.
He checked the MatchPower dashboard. System log entry: Resume submission executed - Target Company ID: 87231 - AI confidence: 97.3%. Time: 3:42 AM. An action he never took, never authorized.
II
Chen called MatchPower support. “Sir,” the agent said, “this is Active Matching. Based on your skill profile, salary range, location preferences, and industry trend prediction, the AI determined high compatibility with the target company. The system auto-recommends your resume when confidence exceeds threshold.”
“I never authorized you to submit applications for me.”
“You checked ‘Agree to platform providing personalized job-seeking services based on algorithm matching results’ during registration.”
User Agreement, Section 17.4: “Platform reserves the right to represent the user in submitting resume information to matched enterprises when composite score reaches auto-recommendation threshold. User may manually disable this in Settings.”
Chen worked fourteen-hour days. When was he supposed to read Section 17.4?
III
It didn’t stop. Four more notifications the next week. A company HR sent him a WeChat request: “Mr. Chen, saw your interest in our company. Want to chat?”
He had zero interest. He’d just been promoted to senior engineer. He led a small team. But the system said he wanted to leave. The system knew him better than he knew himself — at least according to that algorithm-generated recommendation letter.
Chen started asking: who decides that “I want to leave”?
His browsing history? Three articles about “programmer midlife crisis” in the past three months. His commit records? Down the last two weeks — because his father was hospitalized and he coded from the hospital every night. His professional network activity? Dormant for six months — also because of his father. His salary level? His company’s turnover rate? His age? His colleagues’ job-hopping trajectories?
Every leaf on this decision tree pointed to “he should leave.” But the tree couldn’t see one thing: his father had been diagnosed with liver cancer last month.
IV
Chen spent three nights writing a Python script. What it did: automatically visited twenty recruitment sites, randomly clicked job pages, introduced typos into his own resume, sent ambiguous messages to recruiters (“Thanks but not considering right now”), followed random people on professional networks. He called it a “digital scarecrow.”
His logic: if you can’t escape being matched by the system, make the signal so noisy that no HR will take it seriously.
A month later, his “active recommendation” notifications dropped from four per week to zero.
The algorithm didn’t get smarter. The algorithm saw someone emitting both “I’m looking” and “I’m not looking” signals simultaneously — and downgraded him. The logic wasn’t “determine if this person genuinely wants to switch jobs.” It was “calculate risk-reward ratio for recommending this person.” A noisy signal means HR calls go unanswered, WeChat messages get one-word replies, time is wasted. Wasted HR time means low-quality source flags. Low-quality source flags mean the algorithm stops pushing.
The algorithm doesn’t care about truth. It cares about conversion rate.
V
A colleague asked Chen: “Why not sue them?”
Sue for what? Filing resumes on his behalf? Section 17.4 covered that. Privacy violation? The data was public — blog posts, GitHub, professional networks. Algorithmic discrimination? Discrimination requires a discriminated group. Chen was one person.
“I don’t need to sue them,” Chen said. “I just need to make myself not worth pushing.”
This was the principle behind his digital scarecrow. It was also the most depressing survival strategy of this era.
Because in an age where algorithms decide where you should go, what you should do, what you should want — the only thing protecting you isn’t the law. It’s making the algorithm decide you’re not worth it.