Notion's October 13 release is less about flashy new agents and more about plumbing: who can connect what to your workspace, what they can see, and how you pay for the AI. Here's what shipped and what it means for a small team.
Notion released version 3.8 on October 13, titled "Context for your agents." According to Notion's release notes, the core idea is that outside AI tools can now work with your workspace in a controlled way. MCP-compatible agents can read your workspace and update docs and task trackers. Admins get a view of who connected each AI tool, when it was last used, and the ability to revoke one person's access without affecting anyone else.
Notion Agent and Custom Agents can also reach out the other way, with more than 150 MCP connections to outside tools; Notion names HubSpot, Stripe and Cloudflare as examples. The Connections settings were redesigned so you can search by category and preview example workflows, and admins can restrict connections to Notion-verified tools or a hand-picked list.
Connecting AI agents to a workspace full of client notes, pricing sheets and project data is where small teams tend to get nervous, and 3.8 addresses that directly:
Notion also rebuilt synced databases, starting with GitHub, Jira and Salesforce, with live sync health, and added notifications to Microsoft Teams for mentions, comments and access requests.
The most unusual change is on billing. ChatGPT Plus or Pro subscribers can use their subscription for AI usage inside Notion and choose a GPT model. Notion says there is no markup. The option is available on Business and Enterprise plans only, so it does not help teams on Free or Plus plans.
What we could not verify: how this interacts with existing Notion AI credits, whether usage counts against ChatGPT plan limits in a way that would surprise a heavy user, and how it works for teams where several people share one ChatGPT subscription. Check Notion's current pricing page and OpenAI's terms before restructuring anything. We previously covered Notion's AI pricing changes in our Notion AI pricing breakdown; this is a separate wrinkle on top of that.
Seven new models were also added, per the release notes, including GPT 6.1-Sol, GPT-6 Luna, Opus 5.5 and Sonnet 5.5, alongside a "Knowledge Board" that Notion says is based on evaluations of model performance on knowledge-work tasks. Treat vendor-published model rankings as one data point, not a verdict.
MCP connections do not replace trigger-based automation. A Zap or a Make scenario runs the same way every time, which is what you want for invoices, CRM updates and anything where a mistake is expensive. An agent working over MCP is better suited to judgment-heavy, lower-stakes work like drafting, summarizing and triaging. As we noted in our Notion 3.6 coverage, most small teams end up using both: deterministic pipes for critical paths, agents for the fuzzy middle, and a human approving anything irreversible.
What 3.8 changes is the trust question. Before, the hard part of connecting an agent to Notion was knowing what it could touch. Column permissions, per-user revocation and connection allowlists make that answerable. If you are weighing Make versus Zapier for the deterministic side, those decisions are unaffected, but you may now want fewer ad hoc scripts reading your Notion data.
Notion 3.8 is a maturity release. It does not add a headline agent; it makes the existing agent story safer to adopt by letting admins see, limit and revoke what is connected. For small teams that were hesitant to let AI tools into a shared workspace, that is the more useful kind of update. Pilot with a low-risk workspace area first, lock down sensitive columns, and keep deterministic automation on anything that touches money.
Details are drawn from Notion's published 3.8 release notes. Features and plan availability may change; verify on Notion's site before purchasing.
See our full breakdown of the best AI tools for small businesses in 2026 — pricing, use cases, and honest tradeoffs.
Read the AI Tools Guide →