Notion's Custom AI Agents Reviewed

- Notion's Custom Agents launched publicly in late February 2026 and are designed to automate three core tasks: answering repetitive questions from a knowledge base, triaging tickets and to-dos, and generating automatic status updates. - Early testing with the company Ramp demonstrated that a single agent answered 4,000 questions over two weeks, which was estimated to have saved 2,000 hours of employee time. - The feature is part of a larger strategic push called "Notion 3.0," which frames AI agents as a fundamental evolution of the product, comparable to the initial launch or the introduction of databases. - Custom Agents can be triggered by various events, including a set schedule, a new email, a Slack message, or a change within a Notion database. They can then perform actions like creating or editing Notion pages, sending Slack messages, or updating database properties. - Instead of a flat per-seat fee, the new agents will operate on a usage-based model using "Notion Credits." This is a business model shift CEO Ivan Zhao believes could unlock a market ten times the size of traditional SaaS. - The feature is free for Business and Enterprise customers until May 3, 2026, after which it will require the purchase of credits. A dashboard allows administrators to monitor credit usage and set limits to control costs. - Users can select the underlying large language model for their agents, with options including OpenAI's GPT, Anthropic's Claude, and Google's Gemini, or allow Notion to select the best model for a given task automatically. - The custom agents build upon Notion's existing "Q&A" AI feature, which acts like an internal search engine that can answer natural language questions based on information contained within a user's workspace.