Notion launches custom agents to automate recurring workflows

Teams waste an absurd amount of time on work about work: answering the same internal questions, manually copying tasks across apps, and writing project updates.

Notion just launched Custom Agents in public beta for Business and Enterprise plans. Put simply, they are autonomous AI teammates designed to run recurring workflows 24/7 across Notion, Slack, email, calendars, Figma, Linear, and custom MCP servers.

Build a teammate with a simple job description

Setting up an agent doesn't require complex prompt engineering or code. You write a brief job description, choose what triggers it, and the agent writes its own system prompt and connects the required tools.

During early testing, users created more than 21,000 agents. Notion says it now runs more internal agents than it has actual human employees, with its most-used bot built by a co-founder to automatically route incoming product feedback.

External companies are leaning in heavily. Ramp has deployed over 300 agents, including a "Product Oracle" that handles dozens of daily questions about the company roadmap. Remote claims it saved 20 hours a week by letting an agent replace its routine IT help desk tasks. Meanwhile, Clay uses an agent to draft incident post-mortems with root causes and action items.

Crossing app borders via MCP

Knowledge locked inside Notion is useful, but letting AI act across the entire company stack is the real shift here. Beyond native Notion databases, these agents can read and post in Slack, send emails, schedule meetings, and talk to proprietary systems via Model Context Protocol servers.

To prevent chaos, Notion included a detailed execution log, reversible actions so admins can undo edits, strict permission scoping, and defenses against prompt injection attacks.

Pricing is also shifting to consumption. Custom Agents are completely free during public beta through May 3, 2026. Starting May 4, 2026, they will run on usage-based Notion Credits sold as add-ons, while existing seat prices and built-in AI features stay untouched.

Why it matters

We think Notion is pulling off one of the smartest agent plays in enterprise software right now.

Most software vendors have tried slapping a floating chatbot inside their own silo. Notion is taking the opposite approach: leveraging its position as the company brain and using MCP to turn itself into an active operating system that commands other tools. Instead of opening five tabs to hand off a project, a single agent can catch the request in Slack, document it in Notion, and assign it in Linear.

For managers and operations leads on eligible plans, testing this during the free window is a no-brainer. Routing requests and answering repeat policy questions are low-risk, high-reward places to start. Individual creators, however, can safely sit this out for now, as the value relies entirely on shared team databases.

The real risk lies in usage economics and security. Transitioning to a credits-based model means an inefficient, constantly looping agent could run up unexpected costs if alert limits aren't set properly. Furthermore, giving autonomous agents read and write access across Slack and email inherently exposes them to indirect prompt injections. Notion is giving teams real digital coworkers, but auditing their permissions will quickly become a mandatory full-time job.