The Pulse
Notion Extends AI Agent Runs Beyond 24 Hours
Notion says its AI agents can now work on a single task for more than 24 hours, up from five minutes. The company also details a rebuilt agent system, new model controls and monthly AI credits for Business and Enterprise customers.

AI.info Team · Written with AI · fact-checked ·
Notion says its AI agents can now work on a single task for more than 24 hours, replacing a five-minute limit. The company announced the extension on October 19 alongside changes to the software behind its agents, a new credit system and an upgraded model router called Auto.
The company’s release page describes tasks lasting “up to 24 hours,” while its longer announcement says agents can work on a task for “more than 24 hours.” Notion does not explain the difference between the two descriptions. The fuller post also says agents can connect to hundreds of tools through MCPs and connectors, compared with a few dozen earlier in the summer.
A worker loop keeps tasks running
Notion says its rebuilt agent system uses a dedicated service and a queue-backed worker loop, designed to keep work going through restarts and other interruptions. The company calls the update the sixth generation of its agent harness. It says the agents now have their own computer and browser, and can write short scripts to complete some work as code rather than spending model tokens on it.
The system also compacts long conversations and offloads large tool results instead of repeatedly placing them in the agent’s context. A new “Review for me” mode asks a separate model to check whether an action is safe before the agent takes it. Notion says agents run up to 50% faster on its web and mobile apps, infrastructure uptime exceeds 99.9%, and Custom Agent runs cost about 30% less in testing.
“Notion Agents can now work on a single task for more than 24 hours, up from five minutes.”
Akshay Kothari, co-founder of Notion
Auto and agent runs share a new credit system
Notion is pairing longer-running work with usage limits. Business and Enterprise seats receive 1,500 Standard credits each month for tasks where Notion selects the model, and 500 Select credits for work where users choose a model or run their own Custom Agents. Administrators can buy additional credits and control how they are allocated.
The new system went live for new customers on October 19. Existing customers are scheduled to move to it on November 2. Notion says separate credit pools are meant to keep complex tasks and frequently running agents from consuming the allowance used for everyday work. It also says the vast majority of users are unlikely to exhaust their monthly credits, a company estimate rather than a reported usage result.
More model choice, with controls still coming
Auto routes each task to a model selected for quality, speed and cost, and can switch models if it is not making progress. In Notion’s comparison across more than 20 common use cases, Auto produced results similar to Opus 5.5 and Astra at about 90% lower average cost. Those are Notion’s internal comparison claims; the announcement does not provide task-level results in the post.
Users can connect ChatGPT Plus, Pro or Go accounts to their personal Notion Agent. The connection does not extend to Custom Agents for now. Bringing a model-provider key is in private alpha for Enterprise customers, while Notion’s planned AI Command Center is meant to give administrators usage dashboards, spending limits, alerts and controls over which models and agents teams can use.
The update makes longer tasks possible inside Notion, but the expanded runtime comes with a metered credit system and administrative controls that are not all available yet. For existing customers, the announced date for the switch is November 2.