Does Notion Use AI?

Yes, Notion Uses AI — But Most Users Underestimate How Far It Goes

Yes, Notion uses AI — and it has for longer than most people realize. Notion launched Notion AI in November 2022, two weeks before ChatGPT's public debut, as a writing assistant powered by an OpenAI model. What started as a brainstorming and summarizing tool has since grown into something much bigger: a full AI layer woven through the entire workspace.

The challenge for readers is that Notion's AI features are scattered across pricing tiers. On the Free plan ($0/member/month), AI is available only as a trial — things like generating docs or autofilling databases. But once AI adoption crossed 50% of customers, Notion bundled AI into its Business and Enterprise plans at no extra charge, according to CNBC. That adoption curve is striking: only 10–20% of customers paid for AI a year before, growing to 30–40% by early the next year.

So what do you actually get? The core built-in features cover four areas:

• Document generation — drafting, editing, and summarizing content natively inside Notion pages, right where your team already works.
• Database autofill — AI that populates database properties automatically, turning manual data entry into a background task.
• Workspace Q&A — asking questions and getting answers grounded in your own pages, docs, and connected apps.
• Notion Agent — the flagship "AI teammate" that handles multi-step tasks like drafting Q1 OKRs from your Company Goals pages and Jira epics.

Notion Agent is where things get genuinely agentic. It can create and edit pages and databases, read PDFs, CSVs, Excel, Word, PowerPoint, and ZIP files, run calculations and simple code, and produce spreadsheets, slide decks, PDFs, and documents. It operates under the user's own permissions — Notion's documentation states plainly that "Notion Agent has the same permissions you do."

Perhaps most interesting for teams evaluating AI tools: Notion Agent supports multiple models, including Claude, the newest GPT models, Gemini, and Grok, with an "Auto" mode that picks the right model per task. You're not locked into a single vendor's reasoning style.

The scale of this shift shows in the numbers. Notion crossed $500 million in annualized revenue, and customers have reportedly built over 1 million agents since Custom Agents launched in February, per TechCrunch. A practitioner view frames Notion AI as a "retrieval layer" — strongest at working with your own workspace data, weaker at raw reasoning — so it works best when Notion is already your team's operating system.

For readers building agents that reach beyond a single workspace, verified agent skills and MCP connections are what extend tools like Notion's into broader workflows. SkillzBench is building a free directory of benchmarked agent skills plus a free MCP connector so you can see which skills actually deliver — worth joining the waitlist if you're wiring agents into your stack.

What Notion AI Does Well — and Where It Falls Short

Notion AI is at its best when your question starts with "what do we already know?" The most useful framing comes from practitioner Ishan Vats of IV Consulting, who describes Notion AI as a retrieval layer, not a reasoning engine: "It knows your workspace. It is weaker at raw reasoning, stronger at 'where is that and what does it say'" (IV Consulting's hands-on comparison).

That strength shows up in real, multi-step work. Notion's own documentation shows the agent drafting Q1 OKRs by pulling together your Company Goals pages, Jira epics, and business reviews — context no outside chatbot has. It can read PDFs, CSVs, Excel, Word, PowerPoint, and ZIP files, run calculations, and produce spreadsheets and slide decks, all under your own permissions.

The gaps are just as real, though. Per the same practitioner analysis, Notion AI offers no image generation, no voice, and no web search — and it is "not a coding tool." Notion's official docs also list hard boundaries the marketing pages don't emphasize:

• Notion Agent cannot create database automations (Notion help docs)
• It cannot edit comments, share pages, or manage workspace settings
• MCP servers have no trigger capabilities of their own (per Notion's MCP documentation)
• MCP connections are gated to Business and Enterprise plans only

The practical takeaway is a division of labor. Vats recommends: "Claude for the thinking, Notion AI for the workspace memory, with ChatGPT in the mix for anything multimodal." He also flags the failure mode he sees most often: "The most expensive mistake we see is teams forcing one tool to be all three layers." Notion AI's value is conditional — if your team already lives in Notion, the retrieval layer shines; if not, "the value drops sharply."

That layered thinking applies to agent tooling generally, not just chat assistants. As Notion positions itself as an orchestration hub — its Developer Platform integrates external agents like Claude Code, Cursor, and Codex (per TechCrunch) — choosing which skills and connections each agent actually gets becomes a real decision. SkillzBench is building a free directory of security-scanned, benchmarked agent skills plus a free MCP connector to make that layer-by-layer setup easier to get right.

How Agents Extend Notion: MCP Connections and Custom Agents

Notion's AI stops being a writing assistant the moment you connect it to the rest of your stack. With Custom Agents, launched in February, Notion's workspace can now run multi-step background tasks that pull data from external apps and act on it — and customers have reportedly built over 1 million agents since launch, according to TechCrunch's coverage of the announcement (a company figure, not independently verified).

The connective tissue is MCP, the Model Context Protocol. Notion offers 17 pre-configured MCP connections that require no setup — including Linear, Figma, GitHub, HubSpot, Stripe, n8n, Ramp, Sentry, and Mixpanel — plus the option to connect any app via custom MCP servers, per Notion's help documentation. These connections work with every trigger type: scheduled runs, event-triggered runs from Notion or Slack, and manual runs from chat or @mentions.

Governance is built in, which matters for any team adopting agents at work. Each MCP connection is unique to a single Custom Agent and uses the authenticator's credentials — connections are never shared across agents, so a compromised agent can't reach apps it was never granted. Every tool also gets one of three execution modes:

• Run automatically — recommended for read-only tools
• Always ask — the default for write tools, so anything that changes data requires confirmation
• Always allow — for trusted, low-risk operations

Workspace admins must enable custom MCP servers and can restrict members to approved connections. One plan caveat: MCP connections are available on Business and Enterprise tiers only.

The bigger story is the Developer Platform, which positions Notion as an agent orchestration layer rather than a note-taking app. Its components include Workers (a cloud-based, sandboxed environment for running custom code, free through August then credit-based), database sync from any API-accessible source such as Salesforce, Zendesk, or Postgres, and an External Agent API with launch partners Claude Code, Cursor, Codex, and Decagon. CEO Ivan Zhao framed it plainly: "Any data, any tool, any agent — that's the big picture for the Notion Developer Platform." A Notion CLI is available on all plans.

This is the same pattern reshaping productivity tools generally: the workspace AI handles what's inside its walls, and MCP connections plus agent skills extend it outward. That outer layer is exactly where quality questions arise — published research has found roughly 26% of agent skills carry at least one vulnerability, which is why SkillzBench is building a free directory of verified agent skills with security scanning and sandbox benchmarks, plus a free MCP connector so you choose exactly what each agent sees. If Notion's agents automate part of your workflow, browse and test free skills on SkillzBench to extend the rest — with the Lift shown separately, and money never touching scores.

For teams already living in Notion, the retrieval-layer strength of Notion AI combined with metered, credit-based Custom Agents makes the agentic leap practical rather than speculative. The orchestration bet is that your workspace becomes the coordination point for AI work across every tool you use.

Getting Started: A Practical Path from Notion AI to Full Agent Workflows

Getting started with Notion AI is less about flipping a switch and more about matching what you buy to what you'll actually use. On the Free plan, AI exists only as a trial — you can test generating docs or autofilling databases, but that's the ceiling, according to Notion's pricing page. Once AI adoption crossed 50% of customers, Notion bundled AI into Business and Enterprise plans at no extra charge, which is where most teams should start.

Plan choice matters even more once agents enter the picture. A practitioner analysis from IV Consulting describes Notion's model as billing per seat plus a credit meter for agents, with Custom Agents metered separately via credits rather than included in the seat. If your roadmap involves background, multi-step agent work, budget for that metered usage from day one rather than discovering it on your first invoice.

Step two: follow Notion's own governance defaults when you connect external tools. Each MCP connection is unique to a single Custom Agent and uses the authenticator's credentials — connections are never shared across agents. When configuring tools, Notion offers three execution modes, and its defaults are instructive:

• "Run automatically" — recommended for read-only tools where mistakes are cheap
• "Always ask" — the default for write tools, so a human confirms every action
• "Always allow" — reserved for tools you've fully vetted

Starting read-only tools in "always ask" mode and graduating them only after you've watched them work mirrors how Notion itself expects teams to roll out agents. Workspace admins must enable custom MCP servers and can restrict members to approved connections, so put that governance in place before individual users start wiring things up. Note that MCP connections are available on Business and Enterprise plans only.

Step three: extend beyond Notion's 17 pre-configured connections — Amplitude, Figma, GitHub, Linear, Stripe, and others — with custom MCP servers of your own. Notion's MCP documentation covers two auth methods (OAuth and header-based) and confirms MCP tools work with scheduled, event-triggered, and manual runs. Custom servers are also how you plug in agent skills: instruction packages that teach external agents like Claude Code and Cursor new capabilities.

That's where diligence matters. Published research found that roughly 26% of published agent skills carry at least one vulnerability, so grabbing skills at random is a real security risk. This is why we're building SkillzBench: a free, independent directory of agent skills where every version is pinned, security-scanned, and benchmarked, plus a free MCP connector so you can extend any agent with only the skills you've chosen. Everything is free — the directory, the connector, the benchmarks — and the waitlist is open now.

Done well, this progression — right plan, conservative execution modes, then carefully vetted custom connections — turns Notion from a workspace with AI features into the orchestration layer Notion's CEO describes: "any data, any tool, any agent."