ByteDance dismantles Feishu as AI restructures the company—is enterprise services next?

When a company's most profitable business isn't its most famous product, what should it do?

ByteDance gave an answer. Feishu (Lark), this office software earning 3 billion annually and praised by countless users for being easy to use, was split in two: the product team merged into Doubao, and the sales team merged into Volcano Engine. Honestly, my first reaction was surprise, but thinking carefully, this move is quite ruthless.

Feishu has always had a good reputation in the circle. I used it for a year; smooth, great collaboration, more comfortable than DingTalk or WeCom. But the problem is also clear: It's too expensive. For SMEs, annual fees easily reach hundreds of thousands—that money could hire two operators. Among the small cross-border e-commerce teams around me, none use Feishu; it's too luxurious. ByteDance's decision logic is direct: Since Feishu can't make big money no matter how well it's done, why not split it and let AI (Doubao) and Cloud (Volcano Engine) take over.

The logic behind this restructuring isn't "Feishu is failing," but rather that "AI reconstructing companies" is moving from concept to reality.

ByteDance isn't the first to do this. Early in the year, I wrote a post about iFlytek; they were also adjusting organizational structure, pushing AI capabilities toward core businesses. But ByteDance's move is more thorough, directly splitting a mature product. This reveals a signal: AI is no longer just a tool; it is becoming the company's infrastructure.

Last week I wrote a post about AI phones, mentioning that current AI can simulate human dialogue and handle complex scenarios. But that was at the tool level. ByteDance is placing AI at the foundational layer of corporate architecture, letting Doubao take over Feishu's product logic, and letting Volcano Engine's sales team sell bundled packages of Feishu + Doubao + Volcano Engine. It's like before you bought a TV set (Feishu); now you buy a TV box plus content subscription (Doubao + Volcano Engine). The TV set itself might not be expensive, but content and services are the reasons for continuous payment.

I did the math. Feishu's revenue last year was roughly 3 billion, but this number is negligible in ByteDance's overall revenue. ByteDance's estimated total revenue for 2025 exceeded $200 billion; Feishu accounts for less than 0.2%. Moreover, Feishu's customization costs are high; each major client needs individual solutions, squeezing profit margins thin. In contrast, AI products like Doubao have extremely low marginal costs; selling to 1 million users vs. 100 million users costs almost the same. This is the business ByteDance wants.

This restructuring is essentially betting on a future: AI-native office methods will replace traditional OA. Traditional OA interfaces are tables, documents, and approval flows; users need to actively operate them. But AI-native office might mean you say into a microphone, "Organize last week's meeting minutes and distribute to relevant personnel," and it happens automatically. Merging Feishu's product team into Doubao shows ByteDance no longer wants to make "Feishu with AI features," but rather "a company operating system redefined by AI."

The translation tool I use updated a feature late last year: AI auto-summarization of emails. Every day, AI reads customer inquiry emails first and gives me three suggested replies. This saves me about half an hour daily. But compared to ByteDance's move, mine is trivial. ByteDance aims to run the entire company's workflow on AI.

However, this might be trouble for Feishu's existing clients. A friend's company deeply customized Feishu, spending over a million on secondary development. Now Feishu is split; he worries about future compatibility and service issues. ByteDance responded that Feishu products will continue operating, but future updates will definitely lean towards Doubao and Volcano Engine. For SMEs, this might be good, as AI-powered office tools will become cheaper and easier to use. But for large enterprises, especially those who spent heavily on customization, they surely feel uncertain.

My view is that ByteDance's restructuring rings an alarm bell for all enterprise service companies. In recent years, everyone competed on features, experience, and ecosystem. But now AI is here, and the rules of the game have changed. Whoever finds the true integration point between AI and business processes first will lead. Otherwise, like Feishu, no matter how well done, it might be dismantled and reassembled.

As for users like me, I might not need to agonize over Feishu vs. DingTalk anymore. Because ultimately, I might just use an AI assistant that connects my Douyin DMs, handles my emails, writes copy for me, and auto-replies to customers. Whether it's backed by Feishu or Doubao, I don't care.

Action Advice: If you are currently using Feishu, especially small-to-medium teams, start paying attention to Doubao and Volcano Engine trends. In the next six months, ByteDance will likely launch a series of AI-native office tools, possibly much cheaper than Feishu. If you are a large enterprise, keep communication with your account manager to confirm if existing customization schemes will be affected. Don't wait until the restructuring lands to discover your features have been cut.

AI reconstructing companies isn't adding an AI plugin to Feishu, but building a "company operating system" from scratch using AI. ByteDance moved first; now we see if others follow.