Notion Agent

Notion Agent can turn scattered meeting notes into a task database you can actually work from. You describe the outcome; it finds context, creates pages and updates records. A Custom Agent can repeat that job every Friday. For an ops lead on Business, the useful shift is from copying commitments between pages to reviewing the work an agent has prepared.

Existing Business and Enterprise workspaces moved to the new AI credit system on November 2, 2026. That makes the choice of agent, model and funding source part of setting up the workflow. This guide reflects the rules checked on November 4. Notion's credit transition notice confirms the effective date.

Choose Notion Agent for a Request, Custom Agents for a Routine

Notion Agent works when you ask in chat. Custom Agents add their own instructions, access and automatic starting conditions. Think of the first as an assistant sitting beside you, and the second as a recurring job with a written brief and its own keys.

For personal Agent chat, Auto uses Standard credits; choosing an included model yourself uses Select credits. High credit consumption models use workspace Notion credits. Standard and Select balances are separate, reset monthly and do not roll over. Eligible work pauses when its balance runs out unless an owner enables available workspace credits. Actual consumption varies with the task and context. Running an agent through a connected app or the API uses Notion credits; ordinary integration or API use alone does not. AI Meeting Notes starts consuming Standard credits on December 1, 2026, rather than on the November workspace transition date. These are the distinctions in Notion's current AI credit rules.

A new Custom Agent uses its owner's Select credits, including when teammates run it. It then uses the owner's additional Notion credit allowance, followed by shared workspace credits where available and permitted. An owner can fund it directly from a workspace budget instead. Custom Agents created before November 2 keep using shared workspace credits. Check the agent's actual credit source before assuming it moved to your personal allowance. Custom Agent funding rules cover that exception.

For allowances and cost, use Notion pricing and the Custom Agent cost case.

What It Can Read and Change

The practical boundary is the resource and the connection you give it. Workspace search access and permission to change an external app are separate capabilities.

Start from the relevant page, open Agent, and use @ to identify the source and destination. Use All sources to narrow the search. That gives a request such as “Read @Launch Meeting and add explicit commitments to @Tasks” a useful boundary. The Agent capability reference documents these workspace operations.

The current search connector list is:

The AI connector directory is the source for that list. Connectors are useful for finding and summarizing material, and newly indexed content can lag behind the source app.

Some connections also support action. With your Slack account connected, personal Agent can post messages, reply in threads, edit messages it posted and manage reactions as you. It can reach your permitted private channels and direct messages. Workspace approval may also be needed. See the Slack action requirements.

Connected Gmail supports searching, drafting, sending and organizing email, with confirmation for personal Agent write actions. Connected Calendar supports scheduling and event changes within your permissions. Custom Agents also have Mail and Calendar connections for recurring email and scheduling work; the mail and calendar setup guide explains those workflows.

MCP, a standard connection through which an app exposes tools to an agent, adds further integrations. The available actions depend on the server, authentication and tools you enable. Use the Custom Agent MCP setup guide to check the actual tools before promising a cross-app workflow.

Give It Instructions and Memory Before Delegating

Put lasting working rules in instructions, and keep changing project facts in the source databases.

For personal Agent, click its face to open chat, click the face again, then choose Add instructions. Select an existing page or create your own. An example brief for an ops lead:

Use the Tasks database as the source of truth. Preserve existing owners and deadlines. Cite the meeting page behind each proposed task. Leave missing owners and dates blank and flag them for review. Write concise updates for the operations team.

Anyone who can edit that instructions page can affect your Agent's behavior. Keep the working copy private if it contains personal preferences.

Memories hold durable context such as your role, collaborators and preferences. Personal Agent learns them while chatting; it does not automatically build Memories from all your pages or connected apps. Open Settings → Notion AI → Memory to inspect saved Memories and edit or remove them. Personal Memories are private. Custom Agent Memories have a separate switch and are off by default, controlled by people who can edit the agent. Notion's Memories guide explains the distinction.

A reusable skill is useful when you repeatedly want the same report format while directing the work yourself. It does not replace a schedule. The skills guide separates reusable methods from always-on instructions and autonomous Custom Agents.

What Custom Agents Add, and Who Can Create Them

Ordinary workspace members on Business and Enterprise can create Custom Agents by default. Owners can restrict creation in Settings → Notion AI → Controls, including limiting it to owners or selected members and groups. Guests and restricted members cannot create agents. The current AI settings guide describes the workspace controls.

You can start a Custom Agent on a recurring schedule, or on events such as a database page being added, a property changing, a comment arriving or AI Meeting Notes finishing. Slack triggers include a message, reaction or agent mention in configured channels. Mail and Calendar connections add their own event workflows.

The access model deserves attention: sharing a Custom Agent can let someone retrieve or change information they cannot directly open, if the agent has that access. Its permissions do not shrink to match the caller's. Grant the resources required for the job, then share the agent with the people who should benefit from those resources. The Custom Agent permission rules make this explicit.

Worked Setup: Meeting Notes Into Tasks Every Friday

Build a routine that prepares commitments for an ops lead to review. This is a setup recipe you can implement, not a claim that an agent has already run in your workspace.

The manual job has several steps: read notes, identify commitments, check existing tasks, copy owners and dates, and resolve ambiguities. The proposed agent takes the extraction and copying steps. You keep the decision about whether a task is ready to assign. If Friday cleanup currently takes you an hour, use that as your baseline and measure the remaining review time after the pilot.

1. Prepare the Inputs and Destinations

Create or choose these resources:

Add a Review option to Tasks' Status property. Keep notes in page text so the agent can read them. This routine begins after the notes exist; it does not start a recording.

2. Create the Custom Agent

On desktop or web, open Agents in the sidebar, select +, and choose Create blank, or describe the job in the AI setup chat. Name it “Friday Follow-through.” The Custom Agent setup reference covers these controls and recurring triggers.

3. Grant the Required Access

Under Settings → Tools and access, add Meetings with view access, and Tasks and Run Log with edit access. You need Full Access to those destinations to grant the agent more than view permission. Referencing a page in instructions alone does not connect it.

For this first version, leave web access off. The evidence and output both live in the workspace.

4. Paste a Concrete Brief

Replace the resource names below with the actual page references in your workspace:

Read @Meetings entries whose Ready checkbox is checked and whose Meeting date falls within the past seven days. Extract explicit commitments from their page bodies. For each commitment, search @Tasks for the same Source meeting URL and Source action text, and check for an equivalent existing task. Skip existing commitments and leave their records unchanged. Create new tasks with Status set to Review. Copy an owner and due date only when the notes state them unambiguously. Otherwise leave those fields blank and flag the missing information. Put the source meeting URL and the exact action sentence on each task. Add a @Run Log entry listing created task links, skipped duplicates and questions. If there are no eligible commitments, record that outcome. Do not delete or modify existing tasks.

This makes duplicate checking part of the requested behavior. It is not a database uniqueness guarantee. Avoid overlapping runs, and test the same notes twice before trusting the schedule.

5. Test Manually Before Scheduling

Save and publish the agent with automatic triggers off. Click Run with a small set of sample notes. Include a clear commitment, an item missing an owner and an action already in Tasks.

Inspect the resulting task fields and source links. Run it again: the existing commitment should be skipped. Check Activity for missing access or unexpected changes. Correct the instructions if it fails these checks. This follows Notion's manual testing guidance.

6. Set the Schedule and Funding

Open Settings → Triggers, add Recurring, choose weekly on Friday, and select your time and timezone. Check the next scheduled run before saving and enabling it.

Open the agent's Credits control and inspect Credit source. For a team routine, ask the workspace owner to fund it through Workspace budget and set an appropriate limit. An owner can switch from Owner allowance there. Choosing Auto for this Custom Agent does not turn its funding into personal Agent Standard usage.

7. Review the First Friday and Handle Exceptions

Open the Tasks view filtered to Review, resolve missing owners and dates, then move accepted work into your normal queue. Check the Run Log against the source notes and inspect credits used in Notion AI settings.

The seven-day window is a deliberate starting rule. Notes marked Ready after they age out need a manual catch-up run. A missed Friday also needs a wider recovery window. Once the routine is useful, refine that rule before adding more teams or sources.

Six Uses, Ranked by Operational Payoff

These are workflows a team could build from the supported capabilities, rather than measured results.

Start with the first two: their inputs and outputs can remain in Notion, and an operator can inspect whether the job was done well.

Two Services Worth Building Around It

The strongest opportunity is a meeting follow-through setup for teams already using Notion. The buyer is an ops lead whose notes are organized but whose commitments still disappear. DataForSEO's US Google estimate checked for this article puts “meeting minutes template with action items” at 170 monthly searches. That signals interest in structuring follow-through, not 170 ready buyers.

The smallest sellable version could include the Meetings and Tasks structure, one Custom Agent, a review view and a handover that tests duplicates, missing owners and late notes. The catch is differentiation: extraction alone is easy to copy. The useful service is fitting the workflow to how a team actually assigns responsibility and proving that it handles exceptions.

A weekly reporting setup is the broader second opportunity. A founder or agency operations manager could buy an agent that produces a consistent progress, blockers and next-actions report from agreed databases. DataForSEO estimates 390 monthly US Google searches for “weekly status report template.” Both query estimates come from the keyword overview checked on November 4, 2026; neither establishes willingness to pay.

The smallest version could be a report template, named source databases, one scheduled agent and a review step before distribution. The catch is source quality. A polished report can repeat stale project statuses with confidence. Sell the source ownership and review process alongside the automation.

Where the Workflow Still Needs You

Agent cannot start AI Meeting Notes, change page sharing permissions, create reminders or manage workspace billing and member roles. It also cannot create new database automations or templates. Keep those setup tasks with a person.

More fundamentally, the workflow cannot decide whether a vague suggestion was a real commitment. That is why the example preserves the evidence and sends new tasks to review. For work spanning many tools, compare the alternatives in Notion AI vs ChatGPT before making your workspace the center of the process.

What should a meeting action item contain?

Use a concrete action, its source meeting and the named owner and deadline when the notes provide them. In the example workflow, missing ownership or timing becomes a review question rather than an invented assignment.

Can I use a meeting action items template?

Yes. Standardize the notes so commitments are easy to identify, then use matching fields in Tasks. A template makes the input consistent; the Custom Agent supplies the recurring extraction and filing work.

What belongs in a simple weekly status report?

Use completed work, current blockers and next actions, with links to the records behind them. Give the reporting agent that structure and name the databases it should read. A short report with checkable sources is more useful than a long summary of everything it finds.

On Monday, choose one team's meeting database, prepare the Review view and run the workflow manually. Enable Friday only after the output survives a second run on the same notes. Judge it by accepted tasks and review time.

If you want this workflow built around your team's workspace, AI agent development is the next step.