Notion Rebuilds System Around GPT-5 to Enable Autonomous AI Workflows

Notion has moved away from its existing simple prompt-chaining approach and built a new system architecture centered on GPT-5. Through this redesign, AI has evolved beyond a simple assistive tool into an autonomous AI agent that plans and executes on its own.

The new architecture works by having a central Reasoning Model coordinate modular sub-agents. These agents search Notion, Slack, and the web, edit databases, and select and use the tools needed to carry out complex tasks on their own.

Introduced alongside the launch of Notion 3.0, this system, when given a broad task by a user, plans it out, executes it, and reports back the results on its own.

Based on tests using real user workflows, the GPT-5-based system showed the following results:

• A 7.6% improvement over the existing model on outputs based on user feedback
• A 15% performance improvement on difficult Research Mode questions
• A performance improvement of over 100% on multi-step structured tasks such as deadline updates and competitor research

Notion emphasized that when building agent-centric AI, one should evaluate real user tasks rather than synthetic benchmarks, and design the system architecture so that the model can make autonomous judgments.