VentureBeat
Follow
To scale agentic AI, Notion tore down its tech stack and started fresh
Notion rebuilt its productivity software from scratch for version 3.0 to support agentic AI at enterprise scale. This shift moved beyond traditional, step-by-step AI workflows to advanced reasoning models capable of autonomous planning and tool execution. The new architecture allows AI agents to independently select, orchestrate, and run tools across connected environments. This contrasts with older methods that required exhaustive prompting for specific tasks. Notion's approach leverages modular sub-agents that search, query databases, and edit content contextually. The company prioritizes a "better, faster, cheaper" philosophy, balancing latency and accuracy through rigorous evaluation frameworks. This includes identifying and isolating hallucinations by bifurcating evaluation processes. Notion also understands that latency is subjective and depends on the user's need for immediate answers versus exhaustive analysis. The company heavily uses its own product, generating valuable feedback and ensuring models are not regressing. Notion advises other companies to rebuild when foundational capabilities change, treat latency contextually, and ground outputs in trustworthy data. They believe in making hard decisions and staying at the forefront of AI development to create optimal customer experiences.
