置身钉内?

一个方向看似正确的 AI 办公产品,如何在组织目标、商业压力、产品基因和真实用户需求的拉扯中走偏?

ONE 想做“让事找人”的 AI 工作入口,但最终被做成了“让组织更容易看见、追踪、管控人”的工具;它失败的关键不在 AI 技术本身,而在产品定位、组织机制和用户立场错了。

核心观点:
// 产品发心过载
ONE 同时想服务用户、证明钉钉 AI 化、承载无招回归后的战略战役、消化商业化和集团指标。目标太多,导致产品无法保持一个清晰主线。

// To B 的根本矛盾没有解开
钉钉的付费方是老板/管理者,真实高频使用者却是员工。老板要“可控、可视、确定性”,员工要“边界、缓冲、自由度”。ONE 试图同时满足两边,最后实际更偏向管理者。

// AI 放大了钉钉原有的强管控基因
钉钉早期成功依赖“已读未读、强触达、组织确定性”。ONE 把这些能力 AI 化后,用户感受到的不是效率提升,而是更细颗粒度的监控。

// 用户真实反馈被误判
内测和大客户灰度阶段,管理层反馈很好,但基层员工强烈抵触。团队一度把这种抵触理解成“不适应新工具”,而不是产品方向本身有问题,这是文章认为最致命的误判。

// “信息收敛”变成了“信息轰炸”
ONE 原本想用卡片减少消息混乱,但各业务线都往 ONE 接入内容,最后变成“原有钉钉消息 + ONE 卡片”的双重轰炸。

// 发布会倒排工期摧毁了敏捷迭代
项目不是从 MVP、小范围验证、真实数据纠偏开始,而是围绕固定发布会日期高举高打。负面数据出现后,也没有真正修正顶层定位。

// 组织结构让问题被持续放大
产品、设计、研发、商业化、各业务线之间权责不清;业务线为了 AI 指标纷纷接入;一线用户反馈向上传递时被淡化,最终形成“上层看到正反馈,一线承受真实反噬”的断层。

// 它也并非毫无价值
文章承认 ONE 的方向踩中了办公 AI 的趋势:主动式信息流、AI 进入工作流、基于组织上下文做服务,这些都可能是未来方向。失败的是落地方式和用户立场。


无独有偶,硅谷也在反复证明同一个结论:当企业试图用数据和 AI 把人的工作过程彻底量化,最后得到的往往不是效率,而是反感、抵触和讽刺性的“表演生产力”。

Microsoft 的 Productivity Score 是一个很典型的例子。它原本想帮企业理解员工如何使用 Microsoft 365、Teams、Outlook 等工具,提高组织协作效率。但因为它能暴露个人层面的使用数据,很快被外界批评为“员工监控工具”。最后微软不得不调整产品,弱化个人级别可识别数据,改成更聚合的组织洞察。这个案例说明了一件事:员工并不是天然反对效率工具,但他们不接受被这样管理。

Amazon 内部统计 AI 使用量的做法也很讽刺。公司想推动员工使用 AI,于是开始看谁用得多、谁 token 消耗高、谁更积极拥抱新工具。但指标一旦被看见,就会被表演。有人开始为了显得自己更 AI-native 而刷 token、制造调用量。最后据报道,Amazon 撤掉了内部 AI 使用排行榜。企业以为自己在衡量生产力,实际衡量到的是员工表演生产力的能力。

Meta 的鼠标和键盘追踪则更直接。据报道,Meta 的内部项目会在员工工作电脑上记录鼠标移动、点击、键盘操作,甚至偶尔截图,用来训练 AI agent 理解真实办公行为。员工反应很强烈:内部帖子的高赞评论是“这让我非常不舒服,怎么退出?”;公告下最常见反应是愤怒表情。Meta 最初明确表示工作电脑无法 opt out,后来在压力下只开放了非常有限的临时暂停机制。它没有真正取消,但已经暴露出一个事实:当公司把员工的工作过程本身变成训练数据,员工首先感受到的不是效率,而是被采集、被观察、被占有。

这些案例和《置身钉内》里的钉钉 ONE 放在一起看,重点就很清楚了:这套逻辑正在失败。

钉钉 ONE 看见了未读、待办、遗漏和事项流转,却没看见员工需要缓冲和边界。

Microsoft 看见了工具使用行为,却撞上了隐私反弹。

Amazon 看见了 AI 使用量,却催生了刷指标。

Meta 想看见鼠标和键盘背后的工作过程,却激起了员工对数据采集的愤怒。

它们共同说明:工作不是越透明越高效,人也不是越可测量越好管理。一旦企业把“过程数据”误当成“真实贡献”,员工就会开始保护自己:要么关闭入口,要么消极抵抗,要么表演指标,要么彻底不信任系统。

所以,钉钉 ONE 的教训并不孤立。它只是更完整地呈现了一个趋势的终局:当 AI 办公产品从“帮人工作”滑向“帮组织看人”,失败几乎是迟早的事。

隐私不是效率的敌人,自由也不是管理的漏洞。很多时候,它们恰恰是一个工具能否被长期信任、长期使用的前提。

How Nvidia stock is like Apple and what may come next

1/ Both companies have a history before becoming the center of the world.

Apple: 1980

Nvidia: 1999

2/ Right before the new era, world is experiencing shocks.

Apple/Great Financial Crisis: iPhone was introduced in 2007; iPhone 3G was introduced in 2008

Nvidia/Russia-Ukraine war: H100 was revealed in Apr 2022

3/ Both become largest company in first few years of key product launch

Apple: 2011, 3 years after iPhone 3G

Nvidia: 2024,  2 years after H100


What comes next for Apple?

  • Samsung competition
  • iPhone 5 disappoint
  • China 2013 315 boycott
  • iPhone shipment peaks
  • innovation capability questioned
  • new business lines (watch) considered disappointment at first

Storms might be coming for Nvidia.

..

AI to disrupt games?

AI has created chaos in many areas, including the gaming industry.

Google Genie was a case in point a few weeks ago.

TakeTwo (-10%), Roblox (-10%), and Unity (-20%) all down after Google debuts AI Game Creation Tool
byu/Rukuba inValueInvesting

However, I don’t think code alone is what makes a game successful.

Many successful games are like basketball or soccer.

It’s a cultural and social thing.

Shooting is fun but that’s not basketball is all about.

I bet AI can create and update new games easily but it’s the same for sports.

There can be new “sports” coming up – they can be fun to play as well. However, the number of players, the audience, the whole league/industry around a classic sport are the moat.

In the pre-AI era, I don’t think the studio that has top-tier coders is guaranteed to have blockbuster games.

$RBLX $TTWO $TCEHY

Big capex is not longer welcomed

US big tech continue to post higher capex outlook for 2026 and those figures are surprisingly large.

However, you now start to negative reactions.

1/ Their own stocks respond negatively

2/ Nvidia stock, which presumably is a beneficiary for higher capex, hasn’t responded very positively

#Why capex is less welcomed?

1/ It could just be higher inflation across the chain. higher price for infrastructure, power equipment and construction workers etc. Therefore, it’s a less-efficient use of money

2/ Investors don’t see immediate growth. The 2026 growth outlooks, which should be supercharged by already massive capex in 2025, is not impressive enough. Investors fear that marginal incremental growth coming from additional capex looks small, at least in the current year.

Xiaomi smartphone GP may drop 30% given rising memory cost

Some simple calculation:

Xiaomi smartphone GPM was 12.6% in 2024, with 192bn revenue.

Xiaomi sold 1.64 billion smartphones that year.

The GP per handset is about 147 RMB in 2024

Across different smartphone models, memory cost is different, ranging from 50-500 per handset.

But in a nutshell, it’s about 12-18% of BOM.

It’s could be about 150 memory cost per handset for Xiaomi, which is similar to GP per handset.

Then if memory cost is rising 50-100%, the entire GP per handset could be at risk.

To offset, Xiaomi may increase prices for customers.

And as a large customer for memory chips, it may not receive full mark-up immediately.

In the end, maybe 1/3 of the memory cost impact of 120 need to be absorbed by Xiaomi.

Then GP per handset could be more like 100-110 RMB.

And as the price increases, volume could be impacted, plus the RMB appreciation recently (two-thirds of Xiaomi smartphone volume is overseas).

Total impact to Xiaomi smartphone GP could be like 65-75bn, or 25-30% negative impact from 2024 level.

Human brain vs AI model

Human brain has 86 billion neurons, which forms 100 trillion synaptic connections.

That 100 trillion is the first-order proxy of model “weights”.

Currently, SOTA AI models could have ~2 trillion parameters or model weights.

For example, OpenAI’s GPT 5.2 model is estimated to have 1-5 trillion parameters, while GPT 3 has 175 billion parameters. Meta Llama 4 Behemoth (MoE) has nearly 2 trillion parameters.

Thus AI models now are closer to human brains. Only 50x difference.

However, human brains is not just 100 trillion synaptic connections.

  • A synapse isn’t a single scalar. It has multiple properties (strength, short-term plasticity, release probability, receptor composition, timing effects, etc.). So raw physical degrees of freedom per synapse could be >1.

  • Not all synapses are independently controllable. Biology adds constraints and correlations (developmental wiring rules, local learning, neuromodulators, homeostasis). That means the effective independent DoF is likely lower than “#synapses × variables”.

  • The brain has lots of additional state beyond synapses. Neuron membrane potentials, ion channel states, neuromodulator concentrations, glial regulation, oscillations, etc. That adds dynamic DoF that don’t map cleanly to “parameters” the way a static model does.

Another thing need to keep in mind how energy-efficient a human brain is.

A typical adult brain runs on about ~20 W.

How to operate a SOTA model?

ChatGPT gives me this

for a dense FP16 2T model, 32 H200 GPUs is the “it loads and runs” baseline, while 48–64+ GPUs is where you start getting reasonable headroom + throughput, depending on your target context and requests/sec.

So about 40 kw.

That would be about 2000x energy consumption than human brain.

Of course human is not just about brain, so about 400x.


GPT 5.2 estimated parameters

The previous 40x P/S sector was SaaS

Cloud and SaaS received premium valuation from 2020 to 2021.

The outperformance started in 2016 and lasted 5-6 years. Watch that outperformance here: https://cloudindex.bvp.com/

Back then, “rule of 40” is the king of valuation metrics, which means “um of revenue growth and profit margin should equal 40%+”. The higher the better of course.

While market fluctuates, you can find the P/S or revenue multiple in the past.

Here, you see that in 2016 companies at rule of 40 receives ~6-7x current year revenue multiple.

Here, you see that in 2020 companies at rule of 40 receives 17x LTM revenue, or 12.8x forward revenue.

During this 2020-21 period, it’s normal to see 30-40x P/S for hot SaaS companies. I remembered Shopify was 40x P/S.

Looks at these charts from here – the evidence of 30-40x P/S glory days.

We all know what happened next.

 

 

That multiple fell back to ~6x for the regression line in 2022, with Fed raising interest rates. See here for the chart.

 

The multiple has stabilized afterwards, from 2022 till now.

Currently, the valuation (forward revenue multiple) is ~4-5x for 2nd and 3rd quantile companies followed by BVP, including the names like Salesforce, Hubspot, Workday, Nutanix, etc.

BVP has introduced the new the Rule of X to give growth more credit btw.

I think some bubble is brewing now, with AI model companies or even chip companies.

However, investors keeps dancing, expecting that Trump will appoint new Fed Chair this year and the new chair won’t raise rates. Trump wants lower rates, not higher.

Maybe we should see another around of crazy valuation first.

And if SaaS outperformed 5-6 years (2016-2021), maybe AI-related stuff should outperform till 2027/28.

Nvidia to become new android+qualcom

In the smartphone era, Google’s Android and Qualcom chips is a powerful solution for many hardware makers.

Nvidia looks to resemble that status alone in self-driving, with this announcement. The Alpamayo + Thor combo provide self-driving software + chip for cars.

More than that, it looks that Nvidia might also do that in robots.

Obviously, Tesla would be the Apple of this era.

Btw, Nvidia’s evaluation data set includes lidar data  https://huggingface.co/datasets/nvidia/PhysicalAI-Autonomous-Vehicles-NuRec

This dataset has a total of 1727 hours of driving recorded from planned data-collection drives in 25 countries and 2500+ cities. The data captures diverse traffic, weather conditions, obstacles, and pedestrians in the environment. It consists of 310,895 clips that are each 20 seconds long. The sensor data includes multi-camera and LiDAR coverage for all clips, and radar coverage for 163,850 clips.

 

Robots should be capex?

Technically, capex should be more one-off than recurring.

Phone used to be a “capex” item. You won’t buy a phone every other year, before the iPhone era. Apple’s P/E multiple expanded when it transformed the category into a more “recurring business”.

In the early stage of AI training, people spend whatever is needed on chips capex. This is due to the increased performance of AI GPUs and thus the efficiency of training. However, after this growth era, I think this is still more of a “capex” item thus the growth should normalize later. Inference is another thing though.

Robots should be capex ultimately. However, in the initial adoption stage, which hasn’t arrived yet, we should see a growth that makes people forget this is a capex category. Then there will be a period of doubt, like when Buffett purchased Apple. And hopefully, the leading robot company by then can transform robot into a “recurring” category like Apple did for smartphone.

 

What I don’t understand about robotaxi…

For the same destination, Baidu’s Apollo robotaxi in Shenzhen will charge RMB 125 (before coupon) vs RMB 40 on Didi express (affordable tier, before coupon) and RMB 50 for regular taxi.

Didi charges 125 before coupon
Didi Express charges 40 before coupon
Regular taxi charges 50

What’s also interesting is that Baidu’s robotaxi estimates that it will take 79 minutes!

Meanwhile Didi estimates it’s about 31 minutes, which is in-line with other map apps’ estimates.

Baidu robotaxi charges more than 2x the taxi price and takes more than 2x the time…

Well done.


Attaching the breakdown of Baidu robotaxi fare (before coupon)