On February 24, 2026, Notion shipped release 3.3 with a single headline feature: Custom Agents. This is not another AI panel waiting for a prompt. It is an agent you assign a job, give a trigger or a schedule, and let run around the clock. Notion’s framing is blunt: hand repetitive knowledge work to “AI teammates” so the humans keep the judgment calls.
For developers and product teams, the interesting part is not only the feature set but the business-model turn. AI capability is moving from bundled subscription to usage-based billing. Starting May 4, agent runs consume Notion credits — the logic of the API metered bill has arrived in an end-user office suite, at scale.
What Custom Agents Actually Do
The core verb is “assign,” not “chat.” You define a responsibility, a trigger condition, or a schedule, and the agent executes on its own. The official use-case list covers:
- Q&A agents that answer repeated team questions from Notion, Slack, Notion Mail, Notion Calendar, and MCP sources
- Task routing agents that capture requests, convert them into tasks, and assign owners
- Status update agents for standup notes, sprint recaps, and OKR reports
- Email triage and drafting, scheduling across participants, and FigJam-to-Notion conversion
Each agent can pick its underlying model. Model choice arrived in Notion 3.2; 3.3 pushes it down to the individual agent level, which quietly turns model routing into an ordinary setting that non-engineers configure.
Triggers, Schedules, and MCP Reach
An agent’s boundary is defined by its integrations. Beyond the first-party connectors for Slack, Notion Mail, and Notion Calendar, 3.3 reaches Linear, Figma, and HubSpot through MCP, and ships templates for common workflows.
The design point matters: agents do not wait for Notion to build each connector. Point one at an MCP server and it can act inside the team’s existing toolchain. For teams already running internal MCP infrastructure, this is a schedulable execution surface at roughly zero extra plumbing cost.
Permissions and Audit Trails for Admins
What kills enterprise adoption is not dumb agents but agents doing unauthorized things. Notion’s answer is a concrete control set: agents operate under a teammate-style permission model with per-item access scopes, can be disabled at any time, log every run, and make changes reversible. On Business and Enterprise plans, admins also control who can create agents at all.
“Every run is logged” and “changes are reversible” are the two minimum conditions for an agent to graduate from personal toy to team infrastructure. Without them, no amount of automation scenarios survives a security review.
Free Now, Credits Later: The May 4 Cutoff
The most practical information in the release is the billing timeline. Custom Agents run free through May 3, 2026. From May 4 they consume Notion credits, sold as an add-on to Business and Enterprise plans, with admins buying as many or as few credits as they need.
That makes now through early May a roughly ten-week free window. The right move is to stand up the workflows that genuinely deliver value during it and capture the run data, not to start budgeting after metering kicks in — when an exhausted credit balance simply stops the agents.
The Early-Tester Numbers
Notion published beta figures: early testers built 21,000 agents, and Notion itself runs 2,800 internally. Among named cases, Remote estimates 20 hours saved per week with better than 95% triage accuracy and more than a quarter of tickets resolved autonomously; Braintrust saves 20 minutes per daily update; Notion’s own security team clears 6+ hours a week with an agent named Scruff. One tester built 15 agents in a single week.
These are vendor-reported numbers, so discount accordingly. But build velocity is itself the signal: when the barrier is low enough that someone ships 15 agents in a week, the bottleneck moves fast from “can we build it” to “how do we govern it.”
What This Means for Product Teams
Three observations. First, productivity suites are converging on the same answer: agents plus MCP. Notion’s position is unusually strong because it is simultaneously the knowledge base, the task system, and the document entry point — an agent there has more to do than in a standalone chat tool. Second, credit billing turns AI cost from a procurement-level subscription decision into a per-team usage decision; teams that can measure value hold the renewal leverage. Third, governance before scale: use the free window to settle naming conventions, permission scopes, and log-review habits. Doing it now is far cheaper than backfilling it after May.
Sources
- Notion 3.3: Custom Agents — Notion
- Buy and track Notion credits for Custom Agents — Notion Help Center
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.