中文
English

Why does Baidu always wake up early but rush to collect late?

2025-07-17
2025-07-16 11:36

111906276928.jpg2025-07-16 11:36

Why does Baidu always wake up early but rush to collect late?

Ye Xiaochai ©

Ye Xiaochai ©

follow with interest

This article is from the WeChat official account: Ye Xiaochai, the author: Ye Xiaochai, the original title: "Robin Lee's internal letter: Why did Baidu get up early, but catch up with the late collection?", and the title is from: Visual China


Article Summary

Baidu's early morning rush to rush late collection is due to organizational issues: distorted high-level information such as difficult to empathize with users for customized experiences, departmental walls hindering collaboration and causing internal friction, insufficient driving force causing employees to passively respond to key tasks, ineffective evaluations distorting promotion channels and ignoring achievements. The core needs to reform the evaluation system, focusing on collaboration and genuine value measurement.


•   ️   High level distortion: CEOs live in a customized world and cannot understand users' real pain points


•      Departmental Wall: Internal business conflicts and cooperation obstacles result in resource waste


•   ⚙️   Lack of driving force: Grassroots employees lack motivation and efficiency in handling dirty work


•      Elite internal friction: Elite culture promotes infighting, resource mismatch hinders innovation


•      Evaluation failure: distorted upward path, promotion of managers, elimination of practitioners


•      Reform direction: binding strategic core actions, encouraging collaboration, exposing real problems

Although I haven't been at Baidu for long, if you ask me which company has the highest status as a programmer, I would definitely say it's Baidu, because product managers really cry and kneel down to make us change our requirements!






But in recent years, Baidu's development has been very strange, causing many jokes and ultimately earning the reputation of "getting up early to catch up late".




Recently, at the quarterly senior management meeting, Robin Lee, the founder of Baidu, finally couldn't help himself. He made an internal speech with the theme of "Seeking Truth and Being Pragmatic", which was a rare way to systematically reflect on Baidu's organizational culture and also review the gains and losses in business.




The so-called "Baidu woke up early and rushed to the late market" is nothing more than these few things:




1. ERNIE Bot grasped the first mover advantage and then collapsed all the way, almost becoming the laughing stock of the industry;




2. Baidu Waimai once had a market share of 33%, but later it was directly eaten by Ele.me;




3. Baidu Intelligent Driving also got up very early, both on the track and in the initial evaluation stage, it was very good, and last year the collapse of Jiyue Automobile caused a stir.




Everything was right, but it seems that the factory director ultimately attributed the reason to lack of focus, which may have been wrong again




How to put it? Perhaps the factory director is different from the world we see!




Our world is different




This is not a joke, the factory director and the world we see may really be different! As far as I know, in some companies, various departments will conduct in-depth warnings on the CEO's account.




Let's take a practical example: a listed boss, whose hometown is in Chengdu, will feed back too laggy videos every time during the Spring Festival.




And the architects below have already deployed a complete monitoring link for the CEO's account, from when the CEO opened the app, to when he clicked a button, watched a video, and performed an action, it was almost restored like a screen recording.




On this basis, they quickly discovered that there was a problem with a CDN node in Sichuan. The result was that the boss only sent a message in the group, and the problem disappeared the next day, but it is unknown whether the problems of others disappeared or not




As far as I know, some big factory departments will develop customized APP functions for CEOs (no matter how complex the technology is), and the boss is likely to see something different from other users: when users complain about a lot of advertisements, the CEO is very confused, and there is no advertisement. When users complain about too laggy video, the CEO's beauty is flying




How do you want the boss to empathize with the users? How do you want your boss to complain about the products? His experience is really great!




Our world is different, and this world is the Truman world bestowed upon the CEO by middle and high-level officials.




Department Wall




We summarized the failure of AI+management 2B entrepreneurship earlier: the success of a company is difficult to summarize because they may not know why they became successful, but poor management will definitely bring down the entire team!




Baidu, as a company, has largely lost in terms of execution. From a personal micro perspective, there are too many elites and smart people in Baidu!




I remember it was still in 2023, when we were working on a medical AI project. In addition to our self-developed model (LLama), we were also trying various large models in China, including Wenxin and Zhipu.




In fact, Baidu's pre-sales work is very awesome. In order to promote Wenxin, two T8 senior engineers were directly assigned to debug with us every day, and achieved some results.




I used to work for Baidu, so I quickly got closer to the engineers and asked about the iteration status of Wenxin. However, the feedback from the execution team was somewhat pessimistic, which roughly means:




Although Wenxin is the leader internally, there are still many teams in the company using LLama for training, and some teams even find it troublesome to organize their data. There are many things that are easier for Yama than for the little devils to deal with




Later, it seems that Baidu sent a "star" team to discuss strategic cooperation with us. Our boss only asked one question before announcing the termination of the cooperation: we are in the medical AI field, you are in the general model field, and Baidu Health is also in the medical AI application field. In terms of layout, will they continue to work on it?




The person in the front row on the other side seems to have a high rank, but when they say something, it immediately turns cold: of course they will do it, how can we control them




This small matter actually reflects a huge problem: Baidu is really not unified internally, and their internal businesses are fighting against each other, and this situation seems to be very common?




The recruitment culture on Baidu's side is very fond of elites, such as qujing, who are particularly active in internal battles. After all, 'internal stability is the first priority in resisting external threats', so there are too many things to do internally and externally.




In fact, every company has a department wall. Baidu has a department wall, which is not a problem. However, at the company level, it is a big problem to not let the basic goals of departments be mutually exclusive when implementing major strategies.




Where is the driving force




Looking closer and pulling the timeline forward, when I was still at Baidu, it was very difficult to take on the big search business, and their product experience was not acceptable to people.




Firstly, the documentation for the company's most important business is extremely poor, and in order to access traffic, I need to consult with at least 5 people.




I still remember asking A a question back then. A asked me to find B, B asked me to find C, C asked me to find D, and D asked me to find E. In the end, E couldn't solve the problem and asked me to find A.




I worked from morning to night and returned to the starting point!




I really don't want to look for A as it is, so I first changed my female nickname on Baidu Hi, and then changed to a fresh and beautiful girl avatar, and then went to ask A. In the end, I don't know if it's an avatar issue or if it's an increase in knowledge after asking around. Anyway, the problem has been solved




There was another time (before changing the avatar) when I had to access the company's core business search because their documents were really bad and there were many language requirements. I couldn't help but complain: What if the user doesn't know that language? Are you considering adding language types?




Unexpectedly, the other person's words turned me upside down: I won't ask your mentor, I have no obligation to teach you!




That's the attitude they have towards the user experience feedback of their products




In fact, looking back, I can understand that maintainer because all he did was dirty and tiring work. If he did it well, he couldn't get promoted or get a raise. If he did it poorly, he still had a big problem, so he had a resentful look on his face. He didn't have much motivation




On the other hand, there is a returnee with a master's degree and a Beijing elder sister (special money!) in the team. Working in a remote location for a large company is very difficult, and I heard that she often wanders around outside during work. Once, the department head asked her to attend a meeting and couldn't find anyone. She lied and claimed that she was visiting a client, and within 10 minutes, she ran from the cinema to the company, which was really amazing




There are many people like this on Baidu, but just looking at the resume panel, they look very excellent




Information distortion and evaluation failure




In fact, many large companies have similar problems. Bosses live in a Truman world and have no idea where the opinions of various departments are being pulled, let alone how the lower level executives get by.




The distortion of information leads to the boss seeing everything as good, and the misjudgment in turn gives the wrong strategy, step by step wrong, step by step wrong.




Evaluation failure leads to a lack of hope for grassroots employees. If there are people around me who are cheating and playing tricks, then I can only do work that pays the right salary.




But the failure of this evaluation is by no means a grassroots issue! It directly points to the core flaw of the company's "elite system": the distortion of upward mobility and the mismatch of resources.




Here, you can compare the big company disease diagnosis book that exploded on Xiaohongshu with the situation on Baidu: bureaucracy is rampant!




The leader himself doesn't end up, he only knows how to draw big cakes and push the front line with macro narrative logic. The ultimate result is a pattern of you fooling me and me slacking off.




Senior sequences need to tackle tough problems. If they are enthusiastic about "analyzing upper level intentions word for word" and turn a blind eye to "important issues" that will truly affect user experience and business advancement, then the product will definitely be extremely difficult to use!




The most fatal problem here may be: those who have been promoted in management and those who have worked diligently have resigned.




From this perspective, the factory director attributed the problem to "lack of focus", perhaps seeing the surface, but the deeper reason is the underlying mechanism that supports "focus": the upward channel has failed.




Without solving the problem of ineffective information transmission and evaluation, strategic focus cannot be implemented.




conclusion




What Baidu needs may not be a simple focus, but a profound reflection: where should resources flow to?




What is the core strategy of the company? What are the key actions that support these strategies, such as bridging collaboration, optimizing core experiences, and tackling key technologies? The evaluation system must be strongly linked to these key actions.




Does writing poor product documentation lead to significantly high collaboration costs? Then make document quality and collaboration efficiency one of the core indicators for promotion and assessment! Has breaking down departmental walls made a significant contribution? The reward should be much higher than guarding one's own territory.




Next, how to measure true value? We need to break the simple mapping of 'rank=value'.




Can we explore more refined value assessment? For example, pricing projects/tasks in a tiered manner to ensure that behaviors that solve complex problems, take on high risks, and create significant user/business value receive corresponding returns?




Let those who do "infrastructure" and "tackle tough projects" have their value seen, measured, and rewarded, rather than just letting those who do "highlight" projects eat meat.




And then breaking the 'Truman World'? The evaluation system needs to encourage mechanisms for reporting concerns, exposing problems, and solving real problems.




For example, setting up a dedicated channel to reward "problem solvers", protecting voices that raise sharp questions, and using "user/frontline pain point resolution rate" rather than "boss satisfaction" as an important assessment indicator. Make the information pipeline more direct and transparent.




In fact, all of the above problems may seem like things, but they are actually human problems. As long as people are right, everything is right. If entropy increase cannot be solved, then find a way to break the system:




Choose the right person, build a new system, and let the leader handle the situation on their own. They should focus on doing a solid job in the business, rather than making judgments behind their backs. This way, the truth cannot be seen.




I believe the factory director's speech on "Seeking Truth and Pragmatism" is just the beginning, but "truth" and "reality" will not automatically emerge. It requires a profound revolution on "how to measure value" and "how to allocate resources".




What Baidu needs is not only a focus on business, but also a reconstruction of an information transmission system and evaluation system that can truly benefit those who seek truth and pragmatism, flow energy towards strategic priorities, and effectively curb entropy increase.




Finally, I hope Baidu continues to improve!




This article is from the WeChat official account: Ye Xiaochai, author: Ye Xiaochai


This content is the author's independent viewpoint and does not represent the stance of Tiger Sniff. Reproduction without permission is not allowed. Please contact for authorization matters hezuo@huxiu.com

If you have any objections or complaints about this article, please contact tougao@huxiu.com

People who are changing and want to change the world are all on the Tiger Sniff app

Related recommendations



为什么百度总是起个大早,却赶了晚集?

本文来自微信公众号:叶小钗,作者:叶小钗,原文标题:《李彦宏内部信:为什么百度起个大早,却赶了晚集?》,题图来自:视觉中国


文章摘要
百度起大早赶晚集源于组织问题:高层信息失真如定制体验难共情用户,部门墙阻碍协作致内耗,驱动力不足让员工消极应对关键任务,评价失效扭曲升职通道忽略实绩。核心需改革评价体系,聚焦协作与价值真实衡量。

• ️ 高层失真:CEO活在定制世界无法理解用户真实痛点

•  部门墙:内部业务互相冲突合作遇阻致资源浪费

• ⚙️ 驱动力缺失:基层员工处理脏活无激励效率低下

•  精英内耗:精英文化助长内斗资源错配阻碍创新

•  评价失效:上升通道扭曲向上管理者升职实干者淘汰

•  改革方向:绑定战略核心动作鼓励协作暴露真问题

虽然我在百度待的时间不长,但你要问我哪家公司的程序员地位最高,我一定会说是百度,因为产品经理是真的会哭着、“跪着”让我们改需求啊!



但近几年来,百度的发展很奇怪,闹了不少笑话,最后落得个“起个大早赶个晚集”的“美誉”。


近日,在季度高管会上,百度创始人李彦宏终于忍不住了,他做了一场主题为《求真务实》的内部演讲,非常罕见地系统反思了百度的组织文化,也复盘了业务上的得与失。


所谓百度起了个大早、赶了个晚集,无非这几件事:


1. 文心一言掌握先发优势后一路拉垮,几乎成为行业笑柄;


2. 百度外卖的市占率曾经一度达到33%,后来却直接被饿了么吃掉了;


3. 百度智能驾驶也起得很早,无论是赛道还是评价初期都非常好,而去年极越汽车倒闭事件闹得沸沸扬扬。


事情都是对的,只不过厂长最终貌似将原因归纳到了不聚焦,这可能又错了……


怎么说呢?也许厂长与我们看到的世界不一样!


我们的世界不一样


这不是在说笑,厂长与我们看到的世界可能真的不一样!就我所知,在某些公司,各个部门会将CEO的账户做深度预警。


举个实际的例子吧:某上市老板老家在成都,每次过年期间都会反馈视频很卡。


而底下的架构师早已为CEO的账号部署了完善的监控链路埋点,从该CEO打开APP,到他什么时候点击了什么按钮,看了什么视频,做了什么操作,几乎像录屏一样被还原了。


在这个基础下,他们很快就发现四川的某个CDN节点有问题,结果就是:老板仅在群里发了个消息,第二天问题就消失了,但其他人的问题消失与否,不知道……


而据我所知,有的大厂部门会为CEO定制化开发APP功能(无论技术复杂度多高),老板看到的极有可能和其他用户不一样:当用户抱怨广告很多的时候,CEO很疑惑,明明没广告啊、当用户在抱怨视频很卡的时候,CEO美女刷得飞起……


你要让老板怎么共情用户呢?你要让老板如何吐槽产品呢?他的体验是真的很好啊!


我们的世界不一样,而这个世界是中高层干部赋予CEO楚门的世界。


部门墙


在前面我们做AI+管理 2B 创业失败的时候就做过总结:企业的成功难以总结,因为他们可能也不知道自己为什么成了,但管理不行一定会拖垮整个团队!


百度这个公司,在很大程度上输在执行上,从个人这个微观角度去看,百度里的精英与聪明人实在太多了!


我记得还是2023年的样子,当时,我们正在做一个医疗AI的项目,除了自研模型(LLama)以外也在尝试国内各种大模型,包括文心、智谱等。


百度的售前其实挺给力的,为了推广文心,直接调了两个T8的高工跑过来天天跟我们一起调试,也取得了一些效果。


我之前是百度的,因此很快就拉近了与工程师的关系,就问到了文心的迭代情况,而执行层的反馈居然是有些悲观的,大概意思是:


虽然文心在内部是大头,但公司内依旧有很多团队在自己用LLama做训练,甚至有些团队的数据整理起来很烦,有很多阎王好过、小鬼难缠的事……


而后,百度貌似派了一个“明星”团队过来与我们探讨战略合作,我们老板只问了一个问题就宣告合作终止了:我们是做医疗AI的,你们是做通用模型的,百度健康也是做医疗AI应用的,从布局来说,他们会继续做吗?


对面排头的貌似职级不低,但说出来话直接就冷场了:他们当然会做,我们怎么管得着他们……


就这一件小事其实体现出了巨大的问题:百度内部是真的不统一啊,他们内部的业务之间是互相打架的,而且这个情况貌似很常见?


百度这边的招聘文化是很喜欢精英的,如qujing这种“精英”,他们在内部干起仗来特别积极,毕竟“攘外必先安内”嘛,所以内斗内行、外斗外行的事情太多了。


其实哪个公司都有部门墙,百度有部门墙没问题,但在公司层面,如何在落实重大战略的时候不要让部门的基本目标互斥,这是很大的问题。


驱动力在哪


再把视野放近,把时间线往前拉,我还在百度的时候,当时要接大搜业务就特费劲,他们那个产品体验就不是人可以接受的。


首先,公司最重要的业务的文档写得是极烂的,我要接入流量,需要辗转问5个人以上。


我至今还记得,当时问A一个问题,A让我找B、B让我找C、C让我找D、D让我找E,最后E解决不了让我找A。


我TM从早上干到晚上,又回归原点了!


我确实不想原样找A,于是先在百度Hi上换了个女性昵称,再换了个小清新的美女头像,然后才去问A,最后不知道是头像问题,还是问了一圈知识增加了,反正问题解决了……


还有一次(换头像前)也是要接入公司核心业务大搜,因为他们的文档确实很烂,还有很多语言上的要求,我忍不住抱怨了一句:万一接入者不会那个语言怎么办呢,你们考虑增加语言类型吗?


谁知对方一句话就给我干翻了:不会问你导师去,我没义务教你!


我那个去:他们对自己的产品的用户体验反馈就是这个态度啊……


其实回过头来看,我是能理解那个维护者的,因为他干的都是脏活累活,做好了不能升职加薪,做差了锅还很大,所以一脸怨妇样,他也没什么驱动力……


另一方面,团队里面有个海归硕士、北京大姐(特有钱!),大公司异地办公很天高皇帝远,听说她经常上班在外面闲逛,有一次部门老大让她开会找不到人,她谎称自己在拜访客户,10分钟从电影院跑到了公司,这真的很屌……


百度里类似这样的人是很多的,但只看简历面板,他们看起来又是很优秀的……


信息失真、评价失效


其实很多大公司都有类似的问题,老板活在楚门的世界,根本不知道各个部门的意见拉扯到什么地步,更不知道底层执行员工是如何得过且过的。


信息失真导致老板看到的都是好的,反过来的误判又给出了错误的战略,一步错,步步错。


评价失效导致基层员工看不到希望,如果身边都是偷奸耍滑的人,那我就只能干对得起工资的活了。


但这评价失效绝不是基层的问题!它直指公司这套“精英体系”的核心病灶:上升通道的扭曲与资源的错配。


在这里,大家可以对比下之前小红书爆出来的大公司病诊断书,与百度的情况很类似:官僚主义盛行!


Leader自己不下场,只会画大饼,用宏观的叙事逻辑去push一线,最终结果就是你忽悠我,我敷衍你的模式。


资深序列是需要啃硬骨头的,如果他们热衷于“逐字逐句分析上层意图”,对于会真正影响用户体验和业务推进的“重要问题视而不见”,那产品必定巨难用!


而这里最致命的问题可能是:向上管理的升职了,踏实干活的离职了。


从这个角度来说,厂长将问题归因于“不聚焦”,或许看到了表象,但更深层的原因是支撑“聚焦”的底层机制:上升通道失灵了。


不解决信息传递与评价失效问题,战略聚焦就执行不下去。


结语


百度需要的,或许不是简单地聚焦,而是深刻反思:资源到底该流向哪里?


公司的核心战略是什么?支撑这些战略的关键动作,如打通协作、优化核心体验、攻坚关键技术又是什么?评价体系必须与这些关键动作强绑定。


写烂产品文档导致协作成本巨高?那就让文档质量、协作效率成为晋升和考核的核心指标之一!破除部门墙有重大贡献?其奖励应远高于守住自己地盘。


其次是如何衡量真实价值? 需要打破“职级=价值”的简单映射。


是否可以探索更精细化的价值评估?比如对项目/任务进行分级定价,让解决复杂问题、承担高风险、创造显著用户/业务价值的行为获得与之匹配的回报?


让做“基建”“啃硬骨头”的人,其价值能被看见、被衡量、被重奖,而不是只让做“亮点”项目的人吃肉。


然后是打破“楚门世界”? 评价体系需要鼓励报忧、暴露问题、解决真问题的机制。


比如,设立专门奖励“问题终结者”的通道,保护提出尖锐问题的声音,将“用户/一线痛点解决率”而非“老板满意度”作为重要考核指标。让信息管道更直接、更透明。


其实,上述所有的问题看起来是事,其实都是人的问题,只要人对了,什么都对了,如果确实解决不了熵增,那就想办法打破系统


选一个对的人,组建一个新的体系,掌门人自己下场,踏踏实实地做好业务,而不是背着手评价,这样看不见真相。


我相信厂长的《求真务实》演讲只是个开始,但“真”与“实”不会自动浮现,它需要一场关于“如何衡量价值”“如何分配资源”的深刻革命。


百度需要的,不仅是业务上的聚焦,更是重构一套能让“求真务实”者真正受益、让能量流向战略要冲、让熵增得到有效遏制的信息传递系统以及评价体系。


最后,希望百度越来越好!


本文来自微信公众号:叶小钗,作者:叶小钗

本内容为作者独立观点,不代表虎嗅立场。未经允许不得转载,授权事宜请联系hezuo@huxiu.com
如对本稿件有异议或投诉,请联系tougao@huxiu.com
正在改变与想要改变世界的人,都在 虎嗅APP

相关推荐

评论
0/500虎嗅用户社区交流公约
热门评论
神评

这些有问题的人是你招来的,这些影响效率的制度是你拍板的,问题都是别人的唯独不是你的,这个逻辑我给满分。

20小时前·安徽
140

推荐看看《大明王朝1566》,谁在那个位置上,大概率也是同样的行为方式。

14小时前·北京
10

百度的先发优势其实一文不值,因为先发并没有做出突破性的成绩,最后还是要跟别人一样抄openai的方案。

19小时前·江苏
50

作者过于基层,看到的现象挺好,但离问题的本质十万八千里。最后竟然把问题归因到人的身上。
百度的根本问题,在于作为一家靠广告收高利润的公司,根本没有动力去创新和探索,这就是所谓资源的诅咒,去看看那些石油富裕的国家,也是一样。
倒退10年,百度可能是互联网的代名词,现如今,已经沦为路人。百度一下你就知道——给豆包用更合适。

14小时前·北京
40

百度所谓的战略眼光其实就是跟着硅谷走,一旦硅谷有什么大动静,百度就立马跟进... PPT

17小时前·上海
30

文心一言是我最早安装的Ai软件,也是我第1个卸载的ai软件。

15小时前·江苏
20
最新评论

作者说的都是战术上的问题,就跟一个人一样,如果做人不行,做事能成吗?百度从来就没有摆脱过“作恶”的标签,一个公司在社会责任和战略目标上脱离了主流价值,做啥能成?

1分钟前·山东
00

不会创新,内部团队效率低

1小时前·江苏
00

归根结底百度的问题是企业文化出了大问题,企业文化就像毛细血管一样输送营养与新鲜血液给每一个组织,它不仅仅指明方向且把组织像石榴籽一样团结在一起,一旦企业文化失真,组织就会衰败,即使战略无比清晰,方向正确,但落实就像35码的脚穿43码的鞋,其次在于选人用人和资源配置之间的关系,企业文化决定一切!

2小时前·未知
00

这个会真开的话,不应该公示出来。应该过两三年后,大家发觉百度好像变了。然后内部人员说曾经开过这样一次会议!不然,基本没改变!

15小时前·云南
00

好事

16小时前·广东
00

CEO都掌控不了公司了,怎么确保在正确的道路上呢。不求每个CEO像雷军又能掌方向又能做营销,但,百度烂成这样了才开始反思是不是迟了点

16小时前·湖北
10

大姐上班期间逛商场,看来工作量不饱和,还能看电影啥。
我看百度反而没啥问题,反正都是混日子,何不找个舒服地方混。

16小时前·上海
10

不用想,内部烂透了

16小时前·河北
00

先驱不成,就变先烈。

16小时前·上海
10

百度这发展,怎么感觉有点坑爹啊?内部反思能解决问题吗?


Read0
share
Write a Review...
推荐文章
1  /  189