Is Feishu the Best Choice for the AI Era? A Good Solution Today, Definitely Not the Best One Long-Term

AI is this capable now, and a platform-style ecosystem like Feishu’s may find its road getting narrower and narrower. On September 15, Feishu held a big conference and said Feishu plus Doubao Work is the best choice for work in the AI era. In early September, ByteDance had just merged the Feishu and Doubao product teams into a new Doubao product team.

From an enterprise’s point of view, private deployment is a long-term, strong need. Feishu’s general-purpose features let a company get a system up and running quickly, and it really is a good low-cost option in China, arguably the most open ecosystem among today’s enterprise AI software and tools. But this kind of platform ecosystem, as opposed to one customized for each company, came about because development used to be expensive and slow. Without that constraint, companies need customized, private tools more. AI is now strong enough that deploying a system of its own won’t cost a company much extra time or effort, and general-purpose platforms like Feishu may be replaced by more private needs and more customized ways of building products.

Feishu has big limitations. It has instant messaging and docs, and it can do every function, but there are compromises all along the way: you’re stuck with Feishu’s formats and its hundreds of complicated APIs. This time, the features Feishu opened up to AI went from two hundred-some to seven hundred-some, yet every one of them has to go through its interfaces.

Some will say that now Doubao Work is out, there’s no need to learn Feishu Base; just tell Doubao in plain language what features are needed. But domestic agents like Doubao Work aren’t cheap. The basic plan is already over a hundred yuan a month, and the compute allowance basically runs out in a day. For normal use, the monthly spend is no less than for top-tier foreign models, and the capability is worse.

Why do I think a product form like Feishu’s may not be the right fit for some time to come? Because the most fundamental need is still the need for privacy and customization. Companies want to digitize the business, employees want a system that fits the business so they can work more efficiently, and the boss wants to make management strategy intelligent and have a boss’s dashboard. These are real needs. But whether it’s Doubao’s intelligence or Claude on top, it’s still Feishu Base being used to build that platform, a framework adopted out of necessity. It’s the company taking the detour: it has to use Feishu’s or Miaoda’s formats to build its own system.

Feishu’s CEO once gave an example: factories replaced steam engines with electric motors and efficiency didn’t change; only later, when they redesigned the production lines around the production process, did efficiency go up. Putting AI into Feishu Base is just swapping in an electric motor. The production line is still the old one.

The premise here is that building a system of its own is a hassle and expensive for a company. That won’t be the case from now on. A friend of mine who manages suppliers built a website for it, done in half a day. Suppliers fill out forms on it and get paid automatically, and whatever the business needs gets built. No ByteDance servers, no Feishu APIs. Any messaging app works, and cross-border business can all be brought together in this one place.

A company’s needs never change: run things its own private way, take the shortest path, digitize the business, make employees more efficient, and make the boss’s management more efficient. Feishu is a good solution right now, but in the long run, it’s definitely not the best solution. Once development and customization costs drop low enough, it’s bound to face challenges.

EP0111 · 2026-10-01 · 03:02. The picture uses a digital human, and the voice uses my cloned voice. This is an English translation of the approved Chinese script used to create the narration.

AI is so capable now that a platform-style ecosystem like Feishu’s may find its road getting narrower and narrower. On September 15, Feishu held a big conference and said Feishu plus Doubao Work is the best choice for work in the AI era. And in early September, ByteDance had just merged the Feishu and Doubao product teams into a new Doubao product team.

From an enterprise’s point of view, private deployment is a long-term, strong need. The reason companies use Feishu right now is that its general-purpose features let them get a system up and running quickly. And Feishu really is a very good low-cost option in China. Feishu actually isn’t the most closed ecosystem. You could even say it’s the most open one among all the enterprise AI software and tools out there right now. But this kind of platform ecosystem, as opposed to an ecosystem customized for each company, came about because development used to be expensive and slow. Without that constraint, what companies really need are customized, private tools. And now AI is strong enough that deploying a system of their own won’t take a company much extra time or effort. General-purpose platforms like Feishu may well be replaced by more private needs and more customized ways of building products.

Feishu has big limitations. It has instant messaging, it has docs, it can do every function, right? But along the way there are all kinds of compromises. I have to use Feishu’s not-so-pretty formats, and its hundreds of complicated API endpoints or MCP. This time, the features Feishu opened up to AI went from two hundred-some to seven hundred-some, but every one of them has to go through its interfaces.

Some people will say that now that Doubao Work is out, there’s no need to learn Feishu Base anymore. Just tell Doubao in plain language what features you want.

But these domestic agents like Doubao Work actually aren’t cheap. The basic plan is already over a hundred yuan a month, but the compute allowance basically runs out in a day. To use it normally, the monthly spend is no less than for top-tier foreign models, and the capability is worse.

So why do I think a product form like Feishu may not be the right fit for some time to come? Because the most fundamental need is still the need for privacy and customization.

Companies want to digitize their business, employees want a system that fits the business so they can work more efficiently, and the boss wants to make management strategy intelligent and have a boss’s dashboard. These are real needs, right? But whether you add Doubao’s intelligence or add Claude, it’s still using Feishu Base to build that platform. It’s forced into that framework, and it’s definitely a detour. It’s the company taking the detour: the company has to use Feishu’s or Miaoda’s formats to build its own system. Feishu’s CEO once gave an example: factories replaced steam engines with electric motors and efficiency didn’t change; later, when they redesigned the production lines around the production process, efficiency finally went up. Putting AI into Feishu Base is just swapping in an electric motor. The production line is still the old one. The premise of all this is that building a system of its own is a hassle and expensive for a company. But from now on, that won’t be the case.

A friend of mine who manages suppliers built a website for it, and it was done in half a day. Suppliers fill out forms on it and get paid automatically. Whatever the business needs, it gets built. No ByteDance servers, no Feishu APIs. A company that wants WeChat can use WeChat, one that wants Telegram can use Telegram, and cross-border business can all be brought together in this one place.

A company’s needs never change: run things its own private way, take the shortest path, and digitize the business, make employees more efficient, and make the boss’s management more efficient. Feishu is a good solution right now, but in the long run, it’s definitely not the best solution. Once development costs and customization costs drop low enough, it’s bound to face challenges.

That’s it for today. See you tomorrow, bye.