Turn Feishu Tables into AI Colleagues that Join Group Chats—No Need to Open the Sheet

On June 23 Anthropic released Claude Tag, an AI agent that can be added to Slack channels as a team member, learn context, and generate code for 65% of its product teams, illustrating the concept of a “Team Agent” that serves the whole team.

Feishu’s Multi‑dimensional Table Agent

Feishu launched a similar feature called “Multi‑dimensional Table Agent”. It lets an AI agent understand team and individual permissions, retain long‑term context, and stay in the team even if a person leaves. The agent can be invoked via direct chat, added to a group, or @‑mentioned in a table comment. Key similarities with Claude Tag are permission handling, proactive work, and shared team knowledge.

Why a “Desk” Is Needed

The article argues that for an agent to become a true team colleague it needs a “desk” – a place that knows the company’s workflow, permission hierarchy, and can continuously store experience and knowledge. Team collaboration software provides this desk; Slack was Claude Tag’s desk, Feishu uses its multi‑dimensional tables as the desk because tables have become more than spreadsheets, supporting various data types, automation, and lightweight apps.

Agent as a Team‑wide Tool

Tables act as a database and workflow engine (e.g., CRM). Therefore they are a natural “desk” for an AI agent, allowing the agent to read/write data, respect permissions, and trigger actions based on business events.

Three‑Step Setup

Select an existing multi‑dimensional table used by the team.

Click “Create” → “Agent”.

Confirm.

In about ten seconds a table becomes an agent that can answer questions, write back data, enforce permissions, and proactively act.

Capabilities

Supports up to 100 knowledge sources (tables, docs, knowledge base, PDFs, images, etc.) forming a long‑term memory and permanent knowledge base.

Permission levels: use, edit, manage; the agent inherits table permissions and cannot expose data beyond a user’s rights.

Long‑term memory is isolated per user, preventing data leakage.

Triggers on data changes, schedule, or custom conditions, allowing non‑deterministic business automation.

Demo Scenario

The authors created a “Selected Topic Management” table and attached an agent. They added the agent to a group chat, gave the group “use” permission, and demonstrated assigning three pending topics to three colleagues by simply typing a command in the chat. The agent added the tasks to the correct rows with appropriate fields.

They also showed that a user without table access cannot retrieve data through the agent, and the agent respects permission changes.

Impact

With the agent, routine workflows no longer require opening the table; a simple chat command can create tasks, update status, or retrieve information. The agent’s memory retains organizational knowledge even when individuals leave, turning experience into a reusable asset.

The article concludes that the hardest part of enterprise AI adoption is not model capability but the underlying “soil” – a mature collaboration platform with structured data, permission granularity, and workflow integration, which Feishu has built over years.