Notion's Ivan Zhao launches Custom Agents — autonomous AI teammates running background workflows without a Mac Mini

Summary

Notion launched Custom Agents, positioning itself as one of the first multiplayer agent tools built for knowledge work that runs in the cloud without requiring local hardware like a Mac Mini. The agents operate autonomously in the background, connecting to external tools like Slack, email, and APIs to handle repetitive workflows.

Notion's pitch differs from existing agent tools in three ways. The agents require no Mac Mini setup, support team-wide permissions rather than personal tinkering, and integrate with systems teams already use. Ramp, which consolidated half a dozen collaboration tools onto Notion last year, has been an early customer using Custom Agents for sales enablement and bug triage.

Zhao frames Notion's strategy as a shift from point-solution replacement to SaaS stack consolidation. Two years ago the company focused on replacing individual tools. Now it is consolidating the broader stack and making that unified platform the agent orchestration layer. Because Notion does not hardcode business logic into the product, language models can use Notion's flexible building blocks to automate work. The company internally describes this as "let AI do the night shift."

Notion is positioning itself as model-agnostic. Zhao says the company will support Claude, Cursor agents, and state-of-the-art models released on any given day, letting users pick models suited to different tasks. He frames this as being "the Switzerland of agents" because individual labs' models do not integrate well with each other.

Margin pressure from model costs is meaningful but not yet at the 10 percentage points Zhao cited in a Wall Street Journal article last August. He expects model pricing to improve as open-source alternatives mature, and says Notion's ability to route work to different models based on task complexity will help offset API costs. The company's strategy is to let customers access frontier capabilities without managing infrastructure, model selection, or setup complexity.

Notion's core bet is that knowledge work consists largely of repetitive, delegable tasks such as paper-pushing, triage, and synthesis, and that agents can handle these tasks asynchronously while teams focus on higher-value work. The infrastructure play is not about model capability itself, which has been sufficient for a year or two. It is about connecting model capability, permissions, and a usable interface that works for companies of all sizes.