Notion CEO on context, Legos versus clay, and agent oversight
- Source
- video.twimg.com
- Author
- Dan Shipper
- Date
Context
Notion's CEO, Ivan Zhao, told Dan Shipper at Thesis: 2027, the conference run by Shipper's company Every, that AI agents have moved where a work-software company's lasting value sits: away from the screens and workflows it designs and toward the data and its customers keep in it. For the past 15 years, Zhao said, software meant a company choosing one good way to run a workflow and building the interface and buttons for it. Agents now open up both parts. They can write the step-by-step code for whatever workflow a person wants, and they can vibe-code a custom interface to go with it.
What remains, in Zhao's account, is how a business holds its data and what that data means to it, the semantics and context that an needs to do useful work. He called that more durable than either the workflow logic or the interface, and said Notion increasingly sees its job as the "keeper" of that data, opened to agents from anywhere while teams of people keep collaborating on it. He described the product side with an analogy: coding agents hand a non-expert clay, code so fine-grained that anything ambitious falls apart, while Notion wants to offer Lego-like pieces. As an example, he said Notion has opened its interface pattern libraries to agents, which can build views on Notion databases while Notion's permissions, syncing and collaboration keep working underneath.
This is a strategy argument, and the evidence for it so far is Notion's own. In a November 5 post, Zhao wrote that roughly half the pages in Notion are now created by agents and that use of Notion through , the standard way outside agents connect to apps, is up roughly 60 times this year. Those figures show agents using Notion; they do not yet show that holding a company's context protects Notion when an agent can rebuild the app around it. In the audience questions, Zhao said many of the pieces are not yet offered out of the box, and that the goal for the next quarter or two is for a 50-person company to get its context set up through agent-led onboarding. The account here rests on Shipper's post and an automatic transcript of roughly the first 30 minutes, which ends mid-answer.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
- vibe coding — Building software by describing what you want to an AI in plain language and steering the result, instead of writing every line yourself.
Checking sign-in…
Loading comments…




