How to Make Agents Collaborate with Your Team on Projects: What Notion 3.8 Lets You Share

Notion 3.8, released October 13, 2026, lets any agent that supports MCP, such as Claude or ChatGPT, read a Notion workspace and update its docs and task trackers. Admins can see who connected each AI tool, check when it was last used, and revoke one person's access without touching anyone else's, while database owners can choose which properties each person or agent may view or edit. The release notes don't say which plans include these features.

What does Notion 3.8 let an outside agent do?

If your projects already live in Notion, the agent you use for drafting can now open them. Notion's release page, titled "Context for your agents," says a connected agent can read workspace context and update docs and task trackers, and that whichever agent you use, the workspace gives it a shared place to find what it needs. You can switch from one agent to another without briefing the new one from scratch.

Alongside it, admins get a view of every connected AI tool, who connected it and when it was last used, with the option to cut off one person's connection on its own. Database owners get column permissions: Notion's own example is a recruiting team working on candidates without exposing compensation.

Why do you keep re-explaining your studio to every AI tool?

You've probably pasted the same project summary into a chat window more than once this month: the client, the scope, the finishes already approved, the vendor whose lead times slipped. Each tool starts cold, so each answer comes back generic, and you end up spending the time you meant to save.

The fix used to sound like a writing project, and most studios reasonably decided they didn't have the hours for it. You already have the record an agent needs in your project pages, meeting notes and vendor lists. Once you connect the workspace, that record becomes context, and the work left for you is deciding what an agent may read. How to Build a Knowledge Base Your AI Agents Can Use covers what's worth adding once you start.

What should an agent be allowed to read in a design studio's workspace?

Give it the context that makes the answer specific: project briefs, scope and the decisions log, vendor lead times and trade terms, your house style, and the spec standards you'd hand a new designer.

Keep back what would cost you if it surfaced in the wrong draft: fee and margin columns, your negotiated trade discounts, staff pay, and the client access notes a design studio collects without thinking about it, such as gate codes, alarm instructions and where the key is left for the installer.

Once agents are reading, you'll find that notes written for yourself are shorthand: "same as the last project's sconces" means nothing to an agent, and a vendor page from last spring reads as current. Column permissions, as Notion describes them, govern database properties; a fee typed into a page body or a meeting note sits outside them, so move numbers you want hidden into a property before you connect anything. Block an hour a week for the first month to read what the agent drafted from your pages, and fix the page whenever the draft gets something wrong.

Does connecting every agent to one workspace lock you in?

It moves the dependency. You're less tied to any one agent, since each new one reads the same pages, and more tied to the workspace that holds the business. That's a reasonable trade for most studios, and it's worth confirming you can export your workspace before it becomes the place everything lives. If your team is already using AI tools you didn't choose, Shadow AI Is a Provisioning Gap covers how to bring those into the open first.

What this changes for a studio

If you've kept decent notes, what you wrote for yourself can now brief any agent you use, and the new work is deciding what each agent may see. It's the first stage of how we work, a single source of truth that your taste and decisions can be read from, and Context Engineering: Why AI Gives You Generic Answers explains why that record matters more than the model.

This week, before you connect anything, open one active project database and list every column and page an agent shouldn't read. Move any fee or margin you've typed into page text into a property you can hide. Then write a one-page studio brief at the top of the workspace: who your clients are, what your rooms always get right, and which vendors you trust. That page helps a new hire as much as it helps an agent, so it's worth writing whether or not you connect one.

All Context Engineering Briefings