Feishu and WeCom Saw the Same Opportunity

Jiang Wu · 2026-03-30

On March 28, 2026, Feishu (known as Lark outside China), ByteDance's workplace suite, open-sourced lark-cli on GitHub under the MIT license. It covers 11 business domains, including messaging, documents, Base tables, calendar, mail, tasks, and meetings, with more than 200 built-in commands and 19 AI agent skills. Two days later, Tencent's WeCom CLI appeared on GitHub the same way, opening seven capabilities: messaging, schedule, documents, smart sheets, meetings, to-dos, and contacts. It supports Claude Code, Codex, QClaw, and other mainstream agents calling it directly.

The two releases came less than 48 hours apart, and their lists of supported agents largely overlap. This looks less like one company's strategic choice and more like a collective statement from the industry. China's two largest enterprise workplace platforms decided at the same time to hand their core capabilities to AI as command-line tools.

The interface stays, with a different job

The logic behind a CLI is direct: AI doesn't need to click, it needs commands. A visual interface is for people, who have to look, understand, and confirm before acting. For AI all of that is noise, and it would rather call the API.

The interface won't vanish for that reason. More likely it changes from a tool for operating into a tool for auditing. People stop using it for every step and use it to check what the AI did. The authority to confirm is being handed over slowly, not in one go, and it dilutes gradually with the frequency of human involvement. Enterprises still need interfaces because they don't fully trust an agent's judgment yet, and that trust can be built up.

"I opened WeCom" may disappear as a sentence

The core assumption of enterprise software used to be that users need a specific interface and channel to reach their tasks. A CLI loosens that assumption.

The caller no longer has to be a person, or a particular device. Claude Code can call it, so can an agent running on a phone, and so could a workflow bot resident in the cloud some day. Once the channel between user and task stops mattering, "work" shifts from opening a certain piece of software on a certain device to a chain of API calls running somewhere. What device starts the chain matters little. What matters is who holds the right to orchestrate it. For now orchestration still mostly sits with people, and it will most likely move further toward agents. By then the sentence "I opened WeCom" may no longer exist.

The data center is worth more than the entry point

Opening a CLI is an active strategic judgment by Feishu and WeCom. They stop treating "users must open my product" as the moat and treat "the user's data lives with me" as the moat.

The logic is very AWS. When Amazon opened its cloud services to competitors, it looked like a giveaway and in fact made Amazon irreplaceable infrastructure. The more tools run on your data layer, the higher the user's cost of leaving, because what has to move is a whole data system and the agent workflows above it, and no longer a single piece of software.

One worry looks reasonable. If an agent operates Feishu on a user's behalf for a long time while the user almost never opens Feishu, is it still the user's tool? The answer is probably yes, and more so than before. The data is with you and so are the permissions, and even if the interface is fully replaced, you remain a node that can't be bypassed.

Both saw the same thing. In an AI environment the whole user service system may not need a visual entry point, and whoever owns the data center owns the user. Opening up today is confidence in being that center.

Sources

Home · All articles · Topics · 简体中文