Notion's core vision has evolved from being a tool for human productivity to a collaborative workspace for humans and AI agents. The goal is no longer for users to do the work themselves, but to manage a swarm of agents that perform tasks on their behalf, using the same primitives like documents and databases for coordination.
Notion deliberately avoids dependency on a single AI lab, instead integrating a variety of models from different providers, including frontier, open-source, and low-cost Chinese models. This "Switzerland for models" approach provides flexibility and allows them to offer the best-performing or most cost-effective option for any given task.
Notion's AI journey demonstrates a pragmatic, phased approach, starting with a simple, single-step "AI Writer" before tackling more complex challenges. They then built a sophisticated Q&A feature with a semantic index across Notion, Slack, and Google Drive, which laid the groundwork for their full agentic product launched in late 2023.
To make their platform usable by AI, Notion had to treat the agent as a new type of customer and build infrastructure specifically for it. This included creating a new markdown dialect and an API allowing agents to interact with databases via SQLite, as well as rewriting their internal AI harness every six months to keep pace with model evolution.
The co-founder, Simon Last, has not manually written code since summer 2023, instead managing agents to perform development tasks. He believes AI tools can create a 100x or even 1000x productivity gap between engineers who leverage them effectively and those who don't, shifting the role from coder to 'agent manager' and verifier.
Keep pulling the thread on Simon Last.