AI 砍掉的第一批大厂人:高薪,高绩效,高P|深氪The first batch of tech giants to be cut: high salary, high performance, high P|Shenke


AI free token/AI免费积分
图片
图片
访谈|任彩茹 兰杰 彭倩
文|任彩茹
编辑|乔芊 杨轩
来源|36氪未来消费(ID:lslb168)
封面来源|IC photo
图片
“630”减员,
AI是祸首还是替罪羊?

“现在公司有(减员)名单,你在这里面。”5月中的一天,林越被组长叫进会议室,对方开门见山。

林越的第一反应是平静,他早有预料。早在今年三四月,一些互联网公司内部便传出要裁员的风声。开年以来,中国互联网大公司围绕AI提效激进开展的token竞赛、培训会、隐形考核等,无处不在。当所有人都被卷入一场“all in AI”的运动时,“裁员一定会发生”就是大家心照不宣的共识。

但站在HR门口时,他还是迎来了情绪崩溃的瞬间:手开始发抖,犹豫了很长时间,想着怎么开头,怎么调整自己的举止表情。“我再也不想经历这样的事。”

林越月薪2万5,一年前本科毕业,入职携程当后端工程师——当时看,他是极其幸运的一个。互联网招聘红利不再,携程几千份简历只录取不到500人,但他进入的是公司最赚钱的酒店部门,负责为商业化产品写代码。

但现在看,月薪2万5、只有一年经验的初级程序员,不裁他裁谁呢?一是赔偿成本低,二是比起对业务通盘更熟的老员工,新人使用AI的效率往往更低。“有业务经验打底,想用AI做什么,有什么影响,老员工更清楚。”林越说。

斯坦福大学在一篇名为《Canaries in the Coal Mine?》(《煤矿里的金丝雀?》)的论文中,用“金丝雀”比喻刚踏入职场的年轻人。其研究显示,2022年ChatGPT普及以来,最年轻劳动者的就业大幅下降,到2025年9月,22-25岁软件开发者的就业相比其2022年底的峰值下降了近20%。

最近一年,AI让一切更卷了。携程曾是著名的“互联网养老厂”:程序员岗早上10点半上班,午休两小时,下午7点就能准时下班,主APP两周一个迭代。但林越入职没多久,赶上AI Coding能力大爆发,已经卷成一周一个APP迭代,“每天干到10点半”。

但这种节奏加快,并不是因为业务有爆发式增长,“而是因为不找事做,就会变成边缘部门,边缘部门就会被砍。”林越告诉36氪。但最终,他还是没有避免“被砍”的命运。

不过,“斩杀”也可能是无差别的。

苍述完全没想到,自己会是第一批出现在裁员名单上的人。

5月的一个周五,上班前半小时,“部门突然拉了一个all hands(全员会),HR直接宣布了结果,告诉大家有这个事儿。”

来美团之前,苍述是字节的SSP校招生(Super Special Offer),高薪入职,到最后也是组内同级员工中工资最高的一个。跳槽到美团后,组内的核心项目几乎都被放在他手中,今年本该是苍述的晋升节点。

这场裁员潮中,“绩优”、“高P”的保护屏障都失效了。苍述的隔壁组,裁掉的两名员工都在去年取得了“超预期”的绩效评级。裁到最后,苍述所在的小组几乎被全员“端掉”,“这个组名义上还存在,实际已经没人了。”

林越得知自己被裁时,才发现平时常常对接的两位前端工程师,“头像不知道什么时候已经灰了”;美团的一个用户增长大群,原本的数百位成员如今只剩一半左右;阿里巴巴的高德、飞猪等业务,也处在剧烈震荡中。

“630”成了社交媒体上的热词。它是国内AI真正大规模走进互联网职场的第一个季度末。六月底到七旬中旬,既是许多公司人员汰换的惯例时点,也是这场裁员潮中普遍设定的“last day”。

风向标硅谷已经率先在裁员,特征是成批量、规模大。5月,Meta宣布裁员8000人,7000人转岗AI部门,成为硅谷科技公司里最动荡的一家,高管承认“公司士气近20年来最低”;更早时候,亚马逊宣布裁员1.6万白领岗位,把省下的资金投向AI。

2021年上一轮裁员潮发生前,国内互联网大公司疯狂扩张边界,高密度地成立一个又一个新业务,一批人被快速招募起来,又快速抹去。

但今年裁员潮的内在主线并非这样单一。AI提效、大而重的老业务增长乏力或深陷竞争泥潭、投资AI新业务带来的现金压力,在这个时段交织并行。许多被通知离开的人,也很难说得清这些因素孰轻孰重。

《哈萨比斯:谷歌AI之脑》的作者称,如同奥本海默创造了原子弹,却无法控制它的使用,追求真理的科学家们也是“万物的破坏者”:我们的工作、思维方式,甚至生存,都可能被“破坏”。十年前的韩国首尔,AlphaGo带给人类棋手李世石最初的破坏。十年后,从硅谷到北京,这种破坏再次蔓延。

对大公司而言,AI是船票,它指向大模型或AI应用这类新业务。但新业务能不能干成、何时干成,没人说得准。面对不再增长的老业务,大公司不得不在每一个确定的和不确定的方向上,更加坚决地提效、进而裁员。

林越向朋友倾诉裁员遭遇时,被安慰道,“没关系,我们大家都会有这一天,只是你的这天来得更早一些。”但比自我开解更重要的可能是,被AI替代、被大厂裁员后,人们该如何选择、如何行动。

图片
焦虑的高层、
加码的中层、发疯的基层

“以前在字节两个月才能做出来的产品demo,我们现在两个星期就能做出来。”一名前字节产品经理、现AI创业公司高管对36氪说,有了Claud Code、Codex这样的工具后,自己团队现在可以3小时做出demo,一周内完成想法验证。

“一个产品(经理)就像一个CEO。”他说,组织结构可以随之大幅压缩,信息传递的损耗比大厂少太多,完美“熵减”。

当创业公司借助AI快速行动之时,互联网大厂回望自身,是否会觉得自己像迟缓的巨兽?

来自大厂最高层的表态,往往是一个信号。

今年3月,美团CEO王兴在高管沟通会上谈到自己对AI的看法,“AI Agent对我的冲击比ChatGPT更大,AI注定会创造巨大生产力,也一定会对组织、对工作模式带来很大的变化。”

 

那场沟通会结束不久,美团在全公司范围拉了一场线上大会,核心是宣贯“龙虾”的安装与使用,提倡每位同事安装“龙虾”,以及将日常工作尽可能写成可复用的Skill。

会后,在美团核心本地商业从事商家运营的陈宇佳收到通知,需要在每周的周报里加上一个板块,写明自己利用AI做了哪些提效、有什么Skill可以推广到全组和全部门使用。“然后就感受到大家好像拼命在把AI融入自己的工作中。”

4月的一天,一位阿里算法工程师毫无征兆地收到部门上一个月的token消耗排行榜,他以170亿token消耗量赫然位列第一名,被公开表扬。部门老板表示,以后年度KPI、晋升考核都将参考这一排行。但一个月后,新的token消耗排行榜没能如期而至,“可能老板也发现了这种排名方式不靠谱。”

新的规则接踵而至。部门领导很快又提出,员工需要在工作日的早上11点到下午6点上传每小时的“时报”,由Agent上的插件自动记录代码及对话内容,生成工作总结——这意味着员工无法修改自己的时报内容。就在第二天,hr以近乎争吵的姿态,劝阻了该领导的荒谬制度。

类似这样的事情,已经不再让人意外。来自高层的AI焦虑不断下放,中层们层层加码,努力暗示下属,这是一场隐形的汇报竞赛、军备竞赛、淘汰竞赛。

尽管没有强制每个人一定要写Skill,陈宇佳的部门领导还是会密切关注每个下属的token用量,不时询问具体情况,“他也不清楚AI具体能做什么,但他说他不允许我们团队的每个人在这次AI浪潮中落后”。有时在工作结束后的私下聚餐中,大家也会接收到老板暗暗传达的一种危机感,“一定要把AI用起来,否则到时候我想拉你们一把都拉不了。”

阿里某AI Coding产品的一位工程师告诉36氪,集团一些业务的老板会向他们的产品团队提出请求,希望通过增加数据埋点的方式,“让他清楚地看到团队成员每天使用AI的具体轨迹。”

美团的一些中层在接到裁员指标后,甚至会向上递出一份更激进、比例更高的裁员名单——更少的人、更高的AI参与度,某种程度上直接等同于新时代的“管理成绩”。

AI提效成了一个任何业务、任何职能都能去“搞一搞”的事情。但关于AI到底能做什么,要怎么落地实现,一条长长的裂缝始终横亘在基层与管理层之间——各级老板对AI赋予无限美好的期待,基层拼命去实现、却总也触及不到那个设想,最终只能疲惫地“表演”。

江灵在阿里淘天集团做客户运营,她的工作是尽可能拉齐消费者需求和商家供给。在她看来,老板们总是“把AI想得很智能、很简单”。

就拿电商中常见的异常场景“爆单”来说,高层期待通过全量巡检,提前找出所有“爆款”。然而,平台一天的商品量是千万级别,远远超出了现有人力和Token可以负荷的数量,于是只能小范围测试,挑选几十万个商品,由于样本范围太小,命中率往往很低。

“作为一个员工,你没法反驳老板的那种期待,你懂吗?”江灵激愤又无奈。

很多时刻,江灵觉得自己就像一头驴,有鞭子在后面抽。“累不可怕,没有方向和正反馈,才是最可怕的。你就是不停地拉磨,也不知道最后要去哪里。”

“你不能把AI当成一个许愿池用。”一家AI公司的CTO对36氪总结,AI提效有很多前提,基础是数据,但很多公司的数字化本身就没做好;此外,许多流程上的卡点在“人”,是单靠AI无法解决的。

图片
“一代人有一代人的土木”

产品、运营等大厂岗位感受到的还是不确定的焦虑,而程序员只能率先接受被宣判的命运。

百度前端工程师李川第一次被AI能力震惊,是在今年初用到Claude Code。“同样的复杂需求,用国内一些大模型可能需要五到六轮对话,用Claude两三轮就搞定了,且完成得更好。”

他第二次被AI惊艳,是今年4月。中国大模型公司智谱发布GLM-5.1模型,“一是便宜,二是它的能力完全可以作为Claude Code的平替。”

李川当时就意识到自己的饭碗不保。到了5月,他果然出现在“名单”上。

如同一枚硬币的两面,一面是2026年5月,Claude Code母公司Anthropic已经实现470亿美元左右的年化收入(ARR),半年时间涨了四五倍;智谱也在近期冲上万亿市值。

另一面是AI Coding能力的极速成熟,让程序员成为了这轮裁员潮中的重灾区。“各家首当其冲的几乎都是产研团队,尤其是前端开发、测试开发这样的岗位,通常很容易被老板认为含金量不再。”一位互联网公司HR告诉36氪。

2025年,李川以校招生身份进入百度,成为一名前端工程师。一年前参与校招面试时,AI还仅仅扮演着搜索引擎的角色,只能通过简单问答来辅助编程,面试官全程都没有谈及AI。

“前端”是李川理想的职业,因为这是一个所见即所得的工作,代码质量直接体现为产品界面上的每个细节。每到过年,告诉家人“打开百度app,上面那个东西是我做的”,会让他尝到成就感和“工作的意义”。

多年以来,大公司的程序员被泾渭分明地划分为算法、前端、后端、测试等职能,前端对审美、交互等软性能力要求更高,后端则更需要严谨的技术能力。这一行的薪资水平和“鄙视链”,也直接跟“技术含量”挂钩——前端比测试高,但不如算法工程师和后端工程师。

仅仅一年,李川熟悉的一切已经地覆天翻。写代码和改代码的工作被AI大面积接管,程序员的几种职能也模糊了界限。甚至产品经理也能一脚跨进编程的门。

阿里的一个开发部门,今年五月接到部门老板的通知,要求大家暂停所有的非紧急需求,每个团队开发一个Agent,以后针对任何业务需求,都只能由产品同事直接与Agent对接。程序员只能修改Agent,不能碰代码。老板还暗示,到今年十月份,做得好的团队将接替不好的团队来维护Agent。

腾讯CSIG的技术团队研发了一条为公司App修复bug的流水线——由AI修复bug,程序员只需要在bug解决完成后进行检查,点击“确认”按钮,代码就会合入,它的修复准确率目前可以达到50%。

阿里巴巴5月在内部成立了一批全栈小组,让前端、后端和测试工程师都转成“全栈工程师”,成为“超级个体”。6月开始,美团内部也在全面推行前后端开发的合并。

转“全栈”理论上是可行的,但实践起来,却是扒掉一层皮的痛苦过程。

突然被转为全栈工程师的韩之,根本没有太多时间学习,很快就要开始自己的第一个“全栈”项目,前后端开发、测试由她一人包揽。“现在我所有需求都是‘倒排’的,规定几号几号前上线,”她最近工作强度打满,晚上9点手头的活儿还没干完,“我实在太累了”。

但大势不可违。从去年底到今年初,中国几家头部公司都在尽可能地撒钱,推动程序员消耗token,逐渐淘汰“古法编程”。

最高峰时,腾讯CSIG团队成员享受的是2000美金/月的token额度,只要诉求合理、有相应的代码产出,用完还可以申请翻倍提额。token使用量也同步被纳入考核,“当你的用量很低时,你的leader会问你为什么。”因此,一些人会把用不完的token额度借给别人。

多年以来,大厂程序员意味着高薪与光环。他们是互联网公司的基石,“程序员精神”的内涵是开源与分享,是代码的简洁与优雅,是没有杂音的唯成果论,是看到字符在屏幕上跳动时的兴奋感。

但时代变了。几乎每一位受访的程序员都对36氪谈到同样一种感受,“离开AI将无法工作,假如AI‘挂掉’了,我宁愿花大量时间去找新的Codingplan,也不愿意自己去看代码改一改”——再去谈论所谓的“程序员精神”,也显得不合时宜。

李川说,昔日一个优秀程序员的修养是学习和迭代,因为过去几十年编程语言一直在变,不学就跟不上技术前沿。他和朋友们周末去咖啡馆研究新技术也是常有的事,“这个群体本身就挺卷的”。但AI恐怖的迭代速度让人彻底失语。

“如果AI Coding锁死在25年的水平就好了,能抹平我这种一两年资历和七八年资历人的技术水平,同时又不能真正代替人,还有很多‘对话框’之外的事可做。”林越感叹。但技术不会为谁停车,现在他毫不怀疑,程序员的消亡已经是进行时了,“就像珍妮机发明后的纺织工人一样”。

图片
旧增长没了,新赛马启动

当技术给一家公司的效率注入倍数级杠杆,随后发生的事情无外乎两种——同样的人做更多的事,又或者,一家公司不再需要这么多人。

“我们不裁员。”一家软件公司CEO对36氪说,好不容易“调教”出了这些对行业、开发方法富有认知的程序员,每一个都是公司的财富,当AI Coding把编程效率提升了5倍,他要做的不是裁掉4/5的人,而是把业务扩大5倍。

这愿望固然美好,但问题在于,市场还有这么多增量吗?

被裁之前,林越短暂体会到AI写代码的“解放感”,但很快,他反倒变得更忙了。以往,业务对App细节有迭代需求时,总要等排期慢慢来。现在业务的需求越堆越快,无论可不可行、重不重要,都让研发团队“先做出来试试”。

但这些需求在林越看来多少有点“鸡肋”——最小的“banner位”修改一下文案细节,又或是将浮窗广告由“免费取消”改为“积分抵扣”。“产品经理变变这个,变变那个,我们会做AB测试,改后效果能变好的情况真的不多。”

“越没有增长的部门越是all in AI,总要找点新故事讲嘛。”苍述说。他既待过外卖业务,也待过无人机业务,以他的亲身感受,前者卷AI的氛围比后者浓得多。

一位刚在Meta经历了大裁员的Infra工程师告诉36氪,学会压榨AI后,他和同事们以前没时间做的事情,“现在都想做一做”。但如今一大批人离开,留下的同事又开始把那些必要性不高的工作砍掉。

摆在所有人面前的现实是,移动互联网时代跑出的明星产品们,如今都很难再通过“做更多工作”而实质性地推高增长。其中一些公司不仅没有增长,还因激烈的外部竞争而失血严重。

2025年外卖大战几家公司烧了2000亿,将美团的利润与现金流拖入泥潭,这让人均贡献利润原本就低的美团率先进入裁员周期。但换个角度看,美团的业务高度依赖线下履约,AI提效空间跟线上化程度更高的公司比是偏小的,“如果连美团都能通过AI提效减员,那其他公司一定会跟进。它是一个风向标。”一位美团员工说。

传统现金牛业务广告持续萎缩的百度,在阿里内部长期边缘、贡献微薄的飞猪和高德,也都是类似情况。

旧业务的裁员难以避免,那么活水的机会存在吗?

一些管理层在谈起裁员话题时,会告诉员工“公司现在也搞AI,可以试着去找找自己能做的项目。”一位美团员工告诉36氪。近日,美团核心本地商业新成立了AI Transformation部门,主要职能是探索用AI梳理业务的内部流程;此外,不少核心中高层也在亲自带队做AI相关项目。

字节的一位产品经理王岳告诉36氪,他正在内部创业,做一款面向B端客户的AI提效类产品,“公司鼓励大家去做这样的探索”。在立项之初,他们不仅主动抹掉了“设计”和“测试”两个职能,还要向评审会强调,这款产品未来将节省多少人力成本。王岳的另一位同事在开发AI客服的Agent产品,其2026年的OKR就是“帮助公司裁掉xx%的客服”。

如今,这样的项目,在每家大厂都有十余个或数十个小团队在做。“有时候会几个团队做同一个方向,谁跑出来了,公司就集中资源推谁。”——一场新的赛马开始了。

变动的除了业务重心,还有组织形态,比如抹去更多中层管理者。

腾讯从今年起开始推行项目制,弱化管理职级,给负责人恢复专业职级;美团在今年年中盘点时裁掉了一些L9(事业部总监级),还在近期全面取消X1节点(此前最低的一层管理节点),减少管理层级。

图片
就让我们挥别过去

AI的巨浪究竟会把人带往何处,多数人都还没有一个“顿悟时刻”。

离职缓冲期结束前的六月中旬,林越已经在密集推进淘宝、快手和字节的面试。继续“大厂程序员”的职业生涯,仍然是他内心期盼的最优路线。但这些公司的橄榄枝至今没能如愿抛来,“太难了。”林越说。

“找到一份工作是容易的,但一旦从大厂去了中厂、小厂,你就不可能再回到大厂了。”在林越心里,放弃大厂某种程度意味着永久性跌落,他不愿“退而求其次”。

也有人放下“大厂执念”。李川从百度离职后的第三天,便无缝入职了一家初创公司。顺其自然地,他的岗位从过去的“前端工程师”变成了“全栈工程师”。这家公司的主力产品是办公类的AI Agent,还给他涨了薪。

尽管大家都说时代变了、程序员的技能不再可靠,李川仍然有一些“技术憧憬”,希望以技术人员的角色参与到一款被用户喜爱的产品中,而这未必只能在大厂实现。

从阿里离开后,江灵入职了一家老牌汽车公司。她如今的工作内容不必强行与AI关联,不用再每天焦虑“老板的AI任务能不能完成”,当然也不用再“拼命表演”。江灵最近负责的一个项目9月30日才上线,“这些任务落在我的舒适区,时间又宽裕,人真的会身心愉悦许多。”

最近,她所在的部门每每放出招聘岗位,“都有一堆什么阿里的人过来面试,疯了一样地往制造业跑。”

也许程序员群体最终还会留下10%,但苍述不想再找大厂的工作,“去卷成这绝望的10%”。

5月被美团裁员后,他果断踏上了创业道路。AI浪潮前,他就以副业形式尝试自己做点什么。彼时,只是建社群、兜售一些技能,就让他体会过月入十万的滋味。

今年三四月时,苍述社群里的一些“学员”,已经踩着浪潮投身AI创业,“开了自己的公司,招了很多人,我还苦哈哈在这上班,这对吗?”他问自己。

如今,苍述的创业项目是面向海外,围绕罕见病用户的需求开发系统、做独立产品,他也在小红书账号“苍述(戒掉月薪版”与海外社媒上向网友分享进展。主产品之外,他还在多线并行一些小产品,以此保持手感,“一个小工具最多三四天完成,复杂系统可能需要半个月。”——这都远远快于大厂的常规排期节奏。

AI或许是人类有史以来最强的智力杠杆,它可以把个人能力放大N倍,可以支撑大部分初创产品的落地,也可以让每一个好的idea被快速看见并定价。

生于2000年的苍述说,自己是一个注定创业的人,但假如没有这次裁员,他可能不会在此刻行动。“公司帮我做了决定。”

“既往不恋,纵情向前”,这是美团给每位离职员工告别短信中的最后一句话,也是最近许多大厂人离开时会提起的一句话。在AI带来的这场复杂变革里,离开大厂、留在大厂,都无法再延续过去的路径。

短暂的“碎”过之后不是躺倒。转行也好,创业也好,先接受变化的人或许能先看到不一样的世界。

(周鑫雨对本文亦有贡献;应采访对象要求,文中林越、江灵、李川、王岳为化名)

 

图片

图片

Interview | Ren Cairu, Lan Jie, Peng Qian

Text | Ren Cairu

Editor | Qiao Qian, Yang Xuan

Source | 36Kr Future Consumption (ID: lslb168)

Cover Source | IC photo

图片

“630” layoffs:

Is AI the culprit or the scapegoat?

“The company has a list of layoffs now, and you’re on it.” On a day in mid-May, Lin Yue was called into the conference room by his team leader, who got straight to the point.

His first reaction was calm — he had expected this. Rumors of layoffs at internet companies had been circulating since March and April. Since the beginning of the year, Chinese internet giants have been aggressively engaging in token races, training sessions, and implicit assessments around AI-driven efficiency. When everyone is swept up in an “all-in-AI” movement, “layoffs will happen” is an unspoken consensus.

But standing at the HR door, he experienced a moment of emotional breakdown: his hands trembled as he hesitated for a long time, wondering how to begin, how to adjust his posture and expression. “I never want to go through this again.”

Lin Yue earns ¥25,000 per month and graduated from university a year ago. He joined Ctrip as a backend engineer — at the time, he was one of the lucky few. With the recruitment bonus gone, Ctrip received thousands of resumes but accepted fewer than 500 people. However, he worked in the hotels department, which was the most profitable part of the company, writing code for commercialized products.

Looking at it now, a junior programmer earning ¥25,000 with only one year’s experience — who else should be laid off? The compensation cost is low, and compared to seasoned employees who understand the business holistically, newcomers are often less efficient using AI. “With business experience as a foundation, senior employees are clearer about what to do with AI and what its impact will be,” Lin Yue says.

In a paper titled “Canaries in the Coal Mine?” by Stanford University, young workers entering the job market are metaphorically referred to as “canaries.” The study shows that since the popularity of ChatGPT in 2022, the employment of the youngest labor force has significantly decreased. As of September 2025, employment of 22–25-year-old software developers is nearly 20% lower than its peak at the end of 2022.

In the past year, AI has made everything more competitive. Ctrip was once known as the “internet retirement factory”: programmers started work at 10:30 AM, took a two-hour lunch break, and quit at 7 PM, with only one iteration per app every two weeks. But shortly after Lin Yue joined, AI coding capabilities exploded, and the pace became “one app iteration per week,” with him working until 10:30 PM every day.

But this acceleration isn’t due to explosive business growth — “it’s because if you don’t find work to do, you become a marginal department, and marginal departments get cut,” Lin Yue tells 36Kr. Ultimately, he didn’t avoid being “cut.”

However, the “killing” may be indiscriminate.

Cang Shu never expected to be among the first on the layoff list.

On a Friday in May, half an hour before work, “the department suddenly called an all-hands meeting, and HR directly announced the result, telling everyone what was happening.”

Before joining Meituan, Cang Shu was a SSP (Super Special Offer) recruit at ByteDance, receiving a high salary. He ended up as the highest-paid worker at the same level in the team. After jumping to Meituan, nearly all core projects were placed under him, and this year was meant to be his promotion milestone.

In this layoff wave, the protective barriers of “high performance” and “seniority” have failed. In the team next to Cang Shu’s, two employees who received “exceeded expectations” performance ratings last year were laid off. By the end, Cang Shu’s entire sub-team was disbanded — “the team officially exists, but there’s no one left.”

When Lin Yue found out he was laid off, he discovered that two frontend engineers he often coordinated with had “avatars that went gray without him knowing it.” A user growth group at Meituan, once boasting several hundred members, is now down to about half. At Alibaba’s AutoNavi and Fliggy, there is also significant turbulence.

“630” has become a trending topic on social media. It marks the first quarter-end after AI was truly integrated into internet workplaces in China. Between late June and mid-July, this is both a traditional time for personnel turnover in many companies and the deadline for the current layoff wave.

The bellwether Silicon Valley is already laying off en masse. In May, Meta announced an 8,000-person layoff and the transfer of 7,000 people to AI departments, becoming the most unstable tech company in Silicon Valley, with executives admitting “company morale is at its lowest in 20 years.” Earlier, Amazon announced the layoff of 16,000 white-collar jobs, shifting funds saved toward AI.

Before the previous layoff wave in 2021, Chinese internet giants were frantically expanding into new businesses, rapidly hiring people who were quickly discarded.

But the underlying theme of this year’s layoffs is not uniformly single-minded. AI efficiency improvements, stagnation in large legacy businesses, and cash pressure from investing in new AI projects are all intertwined. Many people notified of their departure can’t distinguish which factor is more dominant.

The author of “Hasabiz: The Brain of Google AI” notes that just as Oppenheimer created the atom bomb but couldn’t control its use, truth-seeking scientists are also “destroyers of all things”: our work, ways of thinking, and even survival may be “destroyed.” A decade ago in Seoul, AlphaGo’s first destruction was witnessed by human chess player Lee Sedol. A decade later, from Silicon Valley to Beijing, this destruction spreads again.

For large companies, AI is a ticket — it points to new businesses like large models or AI applications. But no one can guarantee when or if such new businesses will succeed. Faced with non-growing legacy businesses, companies must act more decisively on efficiency and layoffs.

When sharing his layoff experience with friends, Lin Yue was comforted: “It’s okay, we’ll all have this day — it just happened to you earlier.” More importantly than self-consolation is what people should choose and do after being replaced by AI or fired by a big company.

图片

Anxious executives, intensified mid-managers, and panicking employees

“Products that used to take two months at ByteDance can now be done in two weeks.” A former ByteDance product manager and current AI startup executive told 36Kr that with tools like Claude Code and Codex, his team can now produce a demo in 3 hours and validate ideas within a week.

“A product manager is like a CEO,” he said — organizational structure can be drastically compressed, reducing information loss compared to big companies — perfect “entropy reduction.”

While startups move quickly with AI, how do internet giants view themselves — are they slow behemoths?

Statements from top executives often signal the mood.

In March, Meituan CEO Wang Xing shared his views on AI at a management meeting: “AI Agents impact me more than ChatGPT; AI will inevitably create huge productivity and change organizations and work modes.”

Shortly after, Meituan held a company-wide online conference promoting the installation and use of “Lobster” (likely an internal AI tool), encouraging employees to turn routine tasks into reusable Skills.

Afterward, Chen Yujia, who does merchant operations at Meituan, received notice to include a section in weekly reports detailing how she used AI for efficiency and which Skills could be shared — “then I noticed everyone trying to integrate AI into their work.”

One day in April, an Alibaba algorithm engineer unexpectedly received the previous month’s token consumption ranking — he topped it with 17 billion tokens, earning public praise. The boss said future KPIs and promotions would consider this ranking. But a month later, no new ranking appeared — “perhaps the boss realized that ranking method was unreliable.”

New rules follow. Department leaders then required employees to submit hourly “time reports” between 11 AM and 6 PM, with an Agent plugin automatically recording code and interactions to generate summaries — meaning no editing of reports was allowed. The next day, HR intervened with near-argumentative force against this absurd制度.

Such incidents are no longer surprising. Anxiety from top executives trickles down, mid-managers intensify pressure, subtly suggesting that lower-level employees are in a hidden competition to perform well with AI.

Though not mandatory, Chen Yujia’s boss closely monitors token usage per employee and asks about specifics. He fears his team lagging in the AI wave. Sometimes during dinners, he imparts a sense of crisis: “Use AI, or I can’t pull you up later.”

Managers of Alibaba’s AI Coding products receive requests from other departments for data points to track team members’ daily AI trajectories.

After receiving layoff quotas, some Meituan mid-managers even propose more aggressive cuts — fewer people, higher AI involvement — equating to “management success.”

AI efficiency has become something any business or function can “tinker with.” But a wide gap persists between executives’ expectations and grassroots realities — bosses envision limitless AI potential, but employees struggle to match those dreams, often ending in exhaustion.

Jiang Ling does customer operations at Alibaba’s Tmall, aiming to align consumer needs with merchant supply. She feels bosses “imagine AI as too smart and too simple.”

Take the common “order flood” scenario: executives expect full-scale inspection to identify all bestsellers. However, with millions of daily products, manpower and tokens are insufficient — so only hundreds of thousands are tested, yielding low hit rates.

“As an employee, you can’t refute the boss’s expectations,” Jiang Ling laments. She feels like a donkey being whipped without direction or feedback — endlessly milling without knowing the destination.

“Don’t treat AI as a wishing well,” advises an AI company CTO to 36Kr — many companies lack good data or have human bottlenecks AI cannot solve.

图片

“Each generation has its civil engineering”

图片

Baidu frontend engineer Li Chuan first marveled at AI capability early this year using Claude Code: “Complex tasks that might take 5–6 rounds with domestic models require only 2–3 with Claude — and better results.”

His second shock came in April with Zhipu’s GLM-5.1 model: cheaper and fully comparable to Claude Code.

This spelled the end of his job security. By May, he was on the “list.”

One side of the coin: by May 2026, Anthropic (makers of Claude) had annualized revenue (ARR) of $47 billion, up 4–5x in six months; Zhipu similarly surged to trillion-dollar valuation.

The flip side: AI coding maturity has decimated programmer ranks. “The first hit is always R&D teams, especially frontend and test devs — bosses often deem them low-value,” says an HR rep.

In 2025, Li Chuan joined Baidu as a frontend engineer. During interviews, AI was just a search engine aid — no mention of its future impact.

Frontend was his dream — what-you-see-is-what-you-get, where code quality reflects product UI. Telling family “I made that thing on the Baidu app” brought pride.

Programmers were once neatly divided by role (algorithm, frontend, backend, test), with salaries reflecting complexity — frontend > test < algorithm/backend.

Within a year, everything changed. Writing and fixing code are now AI-dominated; functional boundaries blur. Even product managers can dabble in programming.

In one Alibaba department, bosses halted all non-urgent tasks and mandated teams to develop an Agent — product managers would interface directly with it, not humans. Programmers only maintain the Agent, not touch raw code. Poor teams risk being replaced.

Tencent’s CSIG built an AI pipeline to fix bugs — humans just verify success; 50% accuracy suffices.

Alibaba formed “full-stack” groups merging frontend, backend, and test — creating “super individuals.” Meituan is also merging roles.

Going “full stack” is painful — sudden retraining while juggling projects leads to burnout.

But resistance is futile. Since late 2024, top firms pushed programmers to consume tokens aggressively, phasing out archaic coding.

At peak, Tencent CSIG offered $2000/month token allowances, doubling on request. Usage was part of performance review — low users got questioned.

Once, programmers embodied high pay and prestige — their ethos was open-source, elegance, and pure meritocracy. Seeing characters dance on screen was exhilarating.

But times changed. Nearly every interviewee told 36Kr: “Without AI, I can’t work — if it fails, I’d rather spend hours finding a new plan than fix code manually.” “Programmer spirit” is obsolete.

Li Chuan notes that historically, learning new languages was part of the job — weekends in cafes discussing tech were normal. Now, AI iterates so fast, no one catches up.

“If AI coding froze at 2025 level, it might bridge gaps between novices and veterans — but even then, many non-coding duties remain. No, we’re like weavers after the spinning jenny — doomed.”

图片

No old growth, new race horses emerge

When tech multiplies efficiency, companies either do more with the same people or need fewer people.

图片

But does the market have that much room?

Before layoff, Lin Yue saw AI’s liberation, then became busier. Non-critical tweaks (e.g., banner texts, rebate labels) piled up — product managers wanted to try everything, even if ROI was unlikely.

“Departments with no growth are the ones all-in on AI — they need a new story,” says Cang Shu — comparing delivery vs drone units, he notes the latter (growing) weren’t as obsessed with AI tokens.

At Meta, post-AI optimization, coworkers had time for extra projects — until layoffs forced them to cut back to essentials.

Reality: mobile-era products can’t boost growth just by “doing more.” Some (like Meituan) bleed cash due to competition.

Meituan’s $2000B burn in 2025 eroded profit and cash flow — a signal that even low-overhead online firms will follow suit if AI savings don’t offset costs.

Baidu (weakening ads) and Alibaba’s Fliggy/AutoNavi (marginal profits) are similar.

So what about new opportunities?

Some managers suggest exploring AI roles during layoffs — Meituan formed an AI Transformation dept and assigned mid-managers to lead projects.

At ByteDance, Wang Yue is building a B2B AI efficiency tool — he intentionally removed design and test roles and emphasizes human cost savings in his pitch. Another colleague builds an AI customer service agent with a KPI to “layoff xx% of customer support.”

Such projects proliferate — multiple teams often tackle the same problem, winner takes all.

Organization also shifts — flattening hierarchies. Tencent now uses project-based roles; Meituan cut L9 directors and eliminated entry-level management nodes (X1).

图片

Farewell to the past

The AI wave’s destination remains unclear to most.

After his notice, Lin Yue applied to Taobao, Kuaishou, Byte — still hoping for a big-tech job — but got no offers. “Too hard,” he says — once out of big tech, it’s hard to return.

Others abandon that dream. Li Chuan left Baidu, joined a startup as a full-stack engineer on an AI office tool — higher salary, more flexible role. He still aspires to be part of a loved product — not necessarily at a giant.

图片

Cang Shu won’t rejoin big tech — he’s now an indie developer for rare disease apps abroad, sharing progress on Weibo and social media. He dabbles in small tools (days to weeks) rather than monolithic systems.

AI amplifies individual potential — Cang Shu might never have left if not forced out.

Meituan’s parting message: “Let go of the past, move forward with gusto.” The era defines the path — whether staying or leaving, the old way is gone.

The world won’t stand still — whether in transition or reinvention, adapting quickly may reveal new vistas.

(Zhou Xinran contributed; names are pseudonyms where requested.)

Although everyone says the era has changed and programmers' skills are no longer reliable, Li Chuan still has some "technical aspirations." He hopes to participate in a product loved by users as a technical person, and this can be achieved even outside of large companies.

After leaving Alibaba, Jiang Ling joined an old-fashioned car company. Her current work doesn't have to be forcibly associated with AI; she doesn't have to worry about whether she can complete the boss's AI tasks every day, nor does she have to "perform desperately." The project Jiang Ling worked on recently was launched on September 30: "These tasks fall within my comfort zone, and the time is generous, so people really feel a lot more comfortable."

Recently, every time her department releases a job opening, "a bunch of people from Alibaba come to interview, rushing into manufacturing."

Perhaps 10% of the programmer group will remain, but Cang Shu doesn't want to look for a job at a big company anymore, "to compete and become part of this desperate 10%."

After being laid off by Meituan in May, he decisively embarked on an entrepreneurial path. Before the AI wave, he had tried to do something on his own as a side hustle. At that time, just building a community and selling some skills allowed him to experience a monthly income of 100,000.

In March and April of this year, some "students" in Cang Shu's community have already jumped into AI entrepreneurship, "started their own companies, hired many people, while I am still working hard here, is this right?" he asked himself.

Now, Cang Shu's entrepreneurial project is oriented toward overseas markets, developing systems for rare disease patients and creating independent products. He also shares progress on the Xiaohongshu account "Cang Shu (Quit the Salary Edition)" and with overseas social media fans. In addition to the main product, he is parallelly working on several small products to maintain his touch, "a small tool can be completed in 3–4 days, and a complex system may take half a month"—this is much faster than the regular scheduling rhythm of big companies.

AI is perhaps the most powerful intellectual leverage humans have ever had; it can amplify individual capabilities by N times, support the landing of most startup products, and allow every good idea to be quickly seen and priced.

Born in 2000, Cang Shu says he is a person destined to start a business, but without this layoff, he might not have taken action at this moment. "The company helped me make the decision."

"Don't cling to the past, move forward with passion," is the last sentence in the farewell message Meituan sends to every departing employee, and it is also a phrase that many big company people mention when leaving recently. In this complex transformation brought by AI, neither leaving big companies nor staying can continue on the past path.

After a brief period of being "broken," one doesn't just lie down. Whether changing careers or starting a business, those who accept the change first may see a different world sooner.

(Zhou Xinyu also contributed to this article; at the request of the interviewees, Lin Yue, Jiang Ling, Li Chuan, and Wang Yue are pseudonyms.)

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry