Guides Notion published apr 9, 2026
This guide teaches you how to build Notion Custom Agents that run recurring company work on a schedule using a simple dual-database system. You will create shared Tasks and Reports databases, then use them with a Weekly Planning agent that turns inbox items into action items and logs its work.
You will build a reusable operating pattern for AI agents: a shared Tasks database for inputs, a shared Reports database for outputs, and a Weekly Planning agent that reads your Gmail inbox every Monday, drafts action items into Tasks, and logs a summary in Reports.
Add a Kanban view on the Tasks database grouped by Assignee so you can see every agent’s work, what is waiting, and what is waiting on you.
Open a new page, add a database, and choose Build with AI. Prompt Notion with:
I need a task database for tracking recurring tasks that agents complete, and for agents to assign one-off tasks to me. Include Name, Type (recurring or one-off), Assignee, Priority, and Status.Notion will create the database and property types for you. The key properties are Type, so agents and humans can filter recurring versus one-off work, and Assignee, so you can assign a task to a specific agent or to yourself.
Pro tip: Add a Kanban view on the Tasks database grouped by Assignee. You will be able to see every agent’s column at a glance, what each one is working on, and what is waiting on you.
Add a second database and prompt Notion to build it:
I need a reports database that teammates and AI agents can add reports to, with a field that marks who (or which agent) created each report. Add views for one-off, daily, weekly, and monthly reports.Reports is the audit trail. Every time a scheduled agent runs, it should add a row here summarizing what it did. This keeps you from digging through Claude project threads or Codex logs to figure out what an agent did yesterday.
Every AI task has inputs and outputs. Tasks are the inputs: what needs to happen. Reports are the outputs: what the agent actually did. Hold every agent to that standard to reduce complexity.
Open Notion AI and click + Create custom agent. Instead of filling fields by hand, prompt it to build the agent for you:
Create a weekly planning agent that looks at my inbox every Monday in Gmail, creates tasks for me in the Tasks database, and then writes a report in the Reports database.
Notion drafts the agent with a name, trigger, instructions, and data sources based on your prompt. Review it, connect the Gmail connector if it is not already wired in, confirm the schedule, and save.
Pro tip: Make sure the agent has edit access to the Tasks and Reports databases.
Anytime you @mention a custom agent in Notion, it triggers the agent to start working.
Open a row in your Tasks database, add the agent as the Assignee, @mention it in a comment, and it goes to work on that task.
Pro tip: You can turn this trigger off in the agent settings on a case-by-case basis.
Stop thinking about AI as a bunch of apps you have to babysit. Start thinking about AI like a business with employees: tasks get assigned, reports get logged, and work gets reviewed on Monday morning.
You do not need Notion Custom Agents for this system to help simplify and automate your business. Any AI that has a Notion connector can plug into the same two databases, including Claude, ChatGPT, Perplexity, Manus, or anything else with scheduled tasks.
Tell the external agent about your Reports database in its system instructions. For example, when you set up a scheduled task in Claude, paste this at the end of the instructions:
When you are done, go to the Reports database in Notion at [paste your Reports database URL] and write a report on what you did today, with Agent = [your agent name] and a one-paragraph summary.Now Claude logs its work in the same place your Notion Custom Agents do. Do the same for every AI tool you use and you will have one dashboard showing what every agent in your stack is doing.
Keep going
Members get all guides, courses, live sessions, and $1,000+ in partner perks.