Start with the databases, not the agents. Build five Notion databases (SOPs, decisions, projects, meeting notes and prompts), give each a named owner and naming rules, then connect agents with scoped access. That workspace becomes the knowledge base inside your AI operating system for a B2B team.

Key points

• Agents are only as good as the pages they read, so database design comes before any agent setup.

• Five databases cover most teams: SOPs, decisions, projects, meeting notes and a prompts library.

• Every database needs one human owner, a status property and a "last reviewed" date.

• Notion's Custom Agents run inside Notion on triggers and schedules; Claude connects from outside through Notion MCP.

• Composio or Relevance AI fit when an agent also has to act in other apps such as Slack or your CRM.

What is a company second brain in Notion?

A company second brain is one shared workspace where the team's procedures, decisions and working knowledge live, written so both people and agents can find them. The idea borrows from personal knowledge management and applies it to a team.

Tiago Forte, who wrote Building a Second Brain, describes a second brain as "an external, centralized, digital repository for the things you learn and the resources from which they come." His PARA method sorts information into Projects, Areas, Resources and Archive, and his CODE method stands for Capture, Organize, Distill and Express.

Forte wrote it for individuals. A company version adds two things he did not need:

• Ownership. Someone is accountable for each database staying current.

• Machine readers. Agents query the same pages, so structure matters more than it does for a human who can skim.

PARA still maps well: Projects become a database, Areas become the SOPs each team owns, Resources hold prompts and reference pages, and Archive becomes a status value.

Which databases does a company second brain need?

Five databases cover most B2B teams under 50 people. Each has a narrow job and a clear answer to "what does an agent do with this?"

Link the databases with relation properties: a meeting note points to a project, a decision to the meeting where it was made. Agents follow those relations the way a new hire would.

Why SOPs come first

SOPs are where an agent gets its instructions. Process Street, which sells SOP and compliance software, frames the goal as turning documents into something people use: "Centralize policies and SOPs into live, searchable workflows to speed onboarding and audits." A Notion SOP database with status and review dates gets part of the way there.

The goal is one canonical page per domain. Fragmented copies of the same doc are where agents pick up stale answers.

How should you name pages and assign ownership?

Use one naming pattern per database and one owner per row, and write both rules on a pinned page agents can read. Consistent names let an agent find the right page instead of guessing between near-duplicates.

Naming rules that hold up:

• SOPs. "Team: Verb + object," such as "Sales: Qualify an inbound demo request."

• Decisions. "YYYY-MM-DD: what was decided," such as "2026-10-01: Annual plans only for new customers."

• Projects. "Quarter: outcome," such as "Q4: Launch partner referral program."

• Meeting notes. "YYYY-MM-DD: meeting name."

• Prompts. "Workflow: step: v#," such as "Inbound reply: draft first response: v3."

Ownership rules:

1. One owner per database. This person sets the properties and approves structural changes.

2. One owner per row. The person property must never be empty on a live page.

3. A review cadence. SOPs and prompts every 90 days is a reasonable starting point; decisions never expire but can be reversed.

4. Retire, do not delete. Set status to Retired so agents skip it but history stays.

Ownership keeps the system alive. A stale pricing page in a database every agent reads makes every agent wrong at the same time.

How do Notion's own agents read and write the workspace?

Notion has two agent features. Notion Agent works with you in the moment, and Custom Agents run on their own in the background.

Notion Agent is the assistant inside the workspace. It creates and edits pages and databases, handles multi-step requests, and follows an instructions page you set, so you can point it at your naming rules.

Notion's help page on Custom Agents says they are "designed to run automatically in the background based on triggers and schedules." Triggers include a page added to a database, a property updated, a comment added, and a finished AI Meeting Note. Slack triggers include a message posted to a channel, an emoji reaction and an @mention of the agent.

Two details matter for a second brain:

• Scoped access. "Agents act only on the pages, databases, and external apps you explicitly grant access to," and Notion says they "never have full workspace access by default."

• Plan requirement. The same page states, "You must be on Notion's Business or Enterprise Plan."

Good first Custom Agents for this setup:

• Meeting follow-up. When an AI Meeting Note finishes, extract action items into Projects and proposed entries into Decisions with status Proposed.

• SOP review reminder. Every Monday, list SOPs past their review date and post the list to the owner's Slack channel.

Give each agent write access to only the database it updates.

Before connecting outside tools, I like to give every agent one master context page with the same ten fields, so each one starts from the same picture of the business.

How do you connect Claude to Notion with MCP?

Use Notion MCP, which lets Claude read and write your Notion pages from Claude's own interface. You connect it once, and Claude can then search the second brain, draft into it and update properties.

MCP is the standard underneath. Anthropic introduced the Model Context Protocol in November 2024 as "a new standard for connecting AI assistants to the systems where data lives," because "every new data source requires its own custom implementation, making truly connected systems difficult to scale."

Notion's MCP help page lists Claude Desktop and Claude.ai among the supported apps and says AI apps can "read from and write to your Notion pages in real-time."

The caveat sits on the same page: MCP tools act with your full Notion permissions, and in Notion's words, "they can access everything you can access." That is a different model from Custom Agents, which start with no access. In practice:

• Check the connecting account's reach so Claude sees the second brain, not every private page.

• Point Claude at the rules page first. Naming patterns, owners and "never edit a Final decision" belong there.

• Keep a person on writes at the start. Claude drafts into Proposed status before anything goes Live.

The guide to rolling out Claude to a team securely covers how to set those access rules before more people connect.

When should you use Composio or Relevance AI instead?

Use them when the agent's job spans Notion and other apps, such as reading an SOP in Notion and then posting in Slack or updating a CRM record. Notion's own agents and Claude via MCP cover Notion-centered work; these tools cover cross-app work with central credential handling.

Composio

Composio connects agents to apps with managed authentication. Its Notion toolkit page lists 57 tools and 8 triggers, including "Create Notion page" and "Query database," which "queries a Notion database to retrieve pages (rows)." Its FAQ says "Composio provides a managed OAuth app for Notion, so you sign in with your normal Notion account and approve access."

Relevance AI

Relevance AI is a platform for building agents without much code. Its Notion integration page lists "33 pre-built Notion tool steps, ready to drop into any agent," including "Get page content as block objects or markdown," plus triggers when a page is created in a data source or a comment is added.

Other tools reading the same pages

Once the second brain is clean, other tools can use it. Databox, for example, offers a Notion MCP connector for its AI Analyst that "reads the docs and roadmap behind your metrics." Form tools such as Paperform can add a page to a Notion database from each submission, which suits project intake.

What does the setup checklist look like?

Build the databases first, then add agents one at a time, so no agent writes into a structure that is still changing.

1. Name a second-brain owner. One person approves database changes and reviews agent logs.

2. Create the five databases with the properties in the table above.

3. Add relations between Meetings, Projects, Decisions and SOPs.

4. Write the rules page. Naming patterns, owners, review cadence and status meanings, pinned at the top of the teamspace.

5. Migrate only live material. Move current SOPs and open projects; leave the rest as archive.

6. Fill the owner and "last reviewed" properties on every live row.

7. Add one Custom Agent with access to one database, such as the meeting follow-up agent writing Proposed entries.

8. Connect Claude through Notion MCP for the people who draft and research.

9. Add Composio or Relevance AI only when a workflow needs another app.

10. Review weekly for a month, then widen access where the output holds up.

For the level of autonomy to grant over time, the L1 to L4 AI automation maturity ladder shows when to move an agent from drafting to acting.

What are the common mistakes when setting up Notion AI agents?

The most common mistake is connecting agents to a messy workspace and expecting them to sort it out. Most others come from missing owners and access that is too wide.

1. Agents before structure. An agent searching thousands of untitled pages returns confident, wrong answers.

2. No status property. Without Draft, Live and Retired, agents cannot tell a current SOP from a 2022 draft.

3. Shared ownership. "The team" owns nothing. Put one person on every row.

4. Wide MCP access by default. MCP acts with the connecting user's full permissions, so connect from an account whose reach you have checked.

5. Letting agents finalize decisions. Agents propose; people mark Final.

6. Prompts kept in chat history. Nobody can review or roll back a prompt that lives in one person's chats.

What does this look like in dollars?

Here is the worked example from my AI Workflow Install page: a 12-person team where each person spends 8 hours a week on manual work, at a $60 loaded hourly rate. That comes to $299,520 a year, and in the example one workflow accounts for half of it, $149,760.

A second brain does not remove those hours by itself. It lets the first automated workflow, and every one after it, work from the same SOPs, decisions and prompts.

To find which workflow to point the agents at first, the manual-work leak audit guide walks through timing each task, converting it to a yearly cost and ranking it. The tools I build this on, including Notion, Claude, Relevance AI and Composio, are listed on my stack page.

FAQ

Do I need a paid Notion plan for AI agents?
For Custom Agents, yes. Notion's help page says you must be on the Business or Enterprise plan.

What is the difference between Notion Agent and Custom Agents?
Notion Agent works with you when you ask it something. Custom Agents run in the background on triggers and schedules you set.

Can Claude edit my Notion databases?
Yes, through Notion MCP. It acts with your full Notion permissions, so check what the connecting account can see first.

How many databases should a company second brain start with?
Five: SOPs, decisions, projects, meeting notes and a prompts library. Add more only when one is clearly overloaded.

Is a company second brain the same as Tiago Forte's Second Brain?
It borrows his ideas, including PARA. Forte's version is personal, while a company version adds owners, review dates and agents as readers.

Should agents be allowed to write to Notion without review?
Not at first. Let them write Proposed or Draft entries, then remove the review step where output is consistently right.

Find out which workflow your second brain should feed first

If you want help deciding which databases and agents your team needs, book a free 30-minute Leak Call. We estimate what manual work in one area costs per year, and you get a costed workflow list within 48 hours. You can also see all my offers on the services page.

About the author: Kevin Payne is an AI engineer and GTM operator who installs AI workflows for B2B teams. He has 10+ years in B2B SaaS demand generation and RevOps across Techstars, a16z and YC portfolio companies, and his writing has appeared in HubSpot, GetResponse, Outbrain and more.