In 2015, Notion was failing. The product had been in development for two years, the team had burned through its initial runway, and growth was flat. Co-founder Ivan Zhao made a decision that looked, from the outside, like surrender: he laid off the team, kept only co-founder Simon Last, and flew to Kyoto to rewrite the entire product from scratch.
The reason for the rewrite was architectural. The existing codebase was built around a fixed set of content types — documents, spreadsheets, task lists — with a rigid data model that could not adapt to how different people and teams actually worked. Zhao’s insight in Kyoto was that the right abstraction was the block: a single, composable unit of content that could be text, image, database, embed, or any future content type, and that users could combine in any configuration.
That block model — conceived in a Kyoto apartment in 2015 — is the architectural decision that made Notion ready for AI in 2023.
Why the 2015 decision unlocked 2023
When Notion launched Notion AI in November 2022, the product’s flexible block architecture meant that AI could be embedded at any level of any document: within a block, across a page, across a workspace. The AI layer did not need to adapt to a rigid content model; the content model was already designed to be extended.
The result: Notion AI adoption crossed 50% of daily active users in under two years from launch — one of the fastest enterprise AI adoption rates any single-product company has publicly reported.
The results — verified sources
- $500 million ARR crossed in September 2025 [CNBC, September 2025]
- 100 million users globally [CNBC Disruptor 50, June 2025]
- Used by teams at over half of the Fortune 500
- Employee tender offer valuing the company at $11 billion in 2025
- Ranked on CNBC Disruptor 50 in both 2025 and 2026
- AI attach rate: above 50% of daily users within two years of AI launch
The pattern: architecture as a long-term asset
The Notion story is unusual because the foundational decision — the Kyoto rebuild — preceded the outcome by eight years. Zhao did not know in 2015 that generative AI would transform the software industry. He knew that the block model was a better abstraction than fixed content types. That judgment turned out to be correct for reasons he could not have fully anticipated.
This is the pattern that appears repeatedly in AI success stories: the companies that extract the most value from AI tend to be the ones that made structural decisions — about data models, about APIs, about composability — years before AI became the dominant topic. The AI integration was made possible by choices that looked unrelated to AI at the time they were made.
What you can take from this
The replicable insight from Notion is not “rewrite your codebase.” It is: the architectural decisions you make today determine what you can integrate tomorrow. Flexibility, composability, and open data models are not just engineering preferences — they are bets on future optionality.
For any product or platform team evaluating architecture decisions today, the Notion question is worth asking: if a capability we can’t yet name becomes available in five years, will our current architecture let us add it? Or will we need to fly to Kyoto?
SAGA — Success Stories & Real Cases · Curating real AI implementations: the original idea, the decision, the verified result.