Notion launches Claude agents in beta to handle busywork inside workspaces

At a glance:

• Notion has launched Claude agents in beta, allowing businesses to assign AI tasks directly inside their workspace.
• The agents can write documents, update pages, generate code, answer questions, and collaborate using shared Notion content, powered by Anthropic's Claude Managed Agents.
• Billing uses Notion credits per run, with no option to connect personal Anthropic accounts, and the feature is off by default for Enterprise and HIPAA workspaces.

How Claude agents work inside Notion

Notion's integration places Claude agents directly within a workspace, turning them into teammates that can act on shared pages, databases, and task boards without users having to copy data between apps. By leveraging Anthropic's Claude Managed Agents, the AI handles the reasoning while Notion manages deployment, security, and the underlying infrastructure, so teams never see separate API keys or backend setup. This tight coupling aims to reduce context-switching and keep all work within a single collaborative environment.

When a user creates a Claude agent from the Agents section, they can start from a template—such as a coding workflow—or build one from scratch with custom instructions, triggers, and integrations. Some templates include GitHub connections that require a personal access token, enabling the agent to read or write code repositories. While the agent works, teammates can follow its progress and inspect the output in real time inside Notion.

Limitations and administrative controls

During any session, Claude agents cannot browse the web or invoke other agents, which means they operate solely on the information already stored in the Notion workspace. This design keeps data private but also limits the agent's ability to pull in external research or live data. Consequently, users must rely on internal documentation for the agent's knowledge base.

Administrative controls give workspace owners the authority to enable or disable Claude agents for their teams. For Enterprise and HIPAA workspaces the feature is disabled by default, and administrators can turn it on via the Notion AI settings page, where they also manage which external users may access specific agents. These controls help organizations meet compliance requirements while experimenting with AI assistance.

Billing, availability and outlook

Instead of charging per API call, Notion bills Claude agent usage through its internal credit system; each run consumes a set number of Notion credits, and users cannot link their own Anthropic account to avoid duplicate billing. Notion takes care of provider access, security patches, and scaling, so organizations can deploy AI agents without managing separate API keys or infrastructure. This approach simplifies procurement and reduces the operational overhead typically associated with third-party AI services.

For companies that already rely on Notion as a central knowledge base, the integration removes a major friction point of modern AI workflows: the constant need to move context between Notion and external chatbots. As the beta expands, watch for broader template libraries, deeper third-party integrations, and potential pricing adjustments based on credit consumption. Early adopters report time savings on routine documentation and code generation tasks, suggesting a productivity boost for teams that embrace the new workflow.