Notion CEO on AI, Company Rebuilding, and the Jazz Band Organizational Model

Linkloud Quote

Notion CEO Ivan Zhao recently recorded a podcast with Sequoia Capital, discussing his experiences of facing two near-fatal company crises and rebuilding from scratch both times. Today, he is applying the same logic to rebuild this thousand-person company, calling himself a “Refounder”:

He believes AI has turned technical skills into commodities; what’s truly scarce is taste and agency, so hiring standards must change; information sharing and coordination tasks are being taken over by AI, so organizational structures must evolve; technological changes happen too rapidly for any planning beyond a few weeks to remain valid, so the way we plan must also change. Enjoy!

In 2015, after two years without finding PMF, with funds running low, Ivan and co-founder Simon made a decision most founders wouldn’t dare to make: lay off everyone, move to Kyoto, and rebuild from scratch. They sublet their San Francisco home and office, and during that time, Notion achieved positive cash flow for the first time.

(Notion's original office in San Francisco)

After settling down, life became extremely simple: write code, eat, write code again, eat again. There was no team, no processes, no resources—just two people and an idea. This experience allowed Ivan to truly understand for the first time that what drives progress is always judgment and determination, while the availability of resources is secondary. One and a half years later, Notion 1.0 was launched.

(Ivan and co-founder Simon’s residence in Kyoto)

The second time was in 2023. The team held an offsite meeting in Cancún, where Ivan gained early access to GPT-4. The experience was a near-shocking revelation for him; he immediately concluded that this would change everything, and that nothing else mattered unless the entire company bet everything on it. He then announced within the 500-person company a restart, fully pivoting toward AI.

But then came nearly a year and a half of hardship. The model technology was still immature, and they tried nearly every direction without success. Growth stalled, and morale plummeted. It wasn’t until the underlying model finally matured that the product took off, with the revenue inflection point and the AI product’s breakthrough occurring almost simultaneously.

In both experiences, what truly mattered was his judgment and his determination to keep moving forward amid uncertainty—qualities that later became the foundation for him to rewrite the company.

Ivan proposed a talent formula:

Talent = Capability × Taste × Agency

Understanding this formula hinges on the derivation process.

Why is Capability depreciating?

Before Google, access to information was a scarce resource, and those who could find information had a real competitive advantage. After Google’s emergence, this advantage disappeared—“I can find this information” became a basic skill. AI is now causing the same phenomenon at the level of capability production. Tasks that once required years of experience to master—such as writing code, crafting copy, or performing data analysis—are now achievable at a decent level by recent graduates using AI tools. The scarcity of capability is being systematically compressed.

Ivan said: "What LLMs accomplish is like how Google enabled everyone to access information—they allow everyone to become decent writers and programmers, giving everyone the capability. But taste still matters; it reflects your value system and what you want to bring to the world. Agency matters too—how hard you work is something no company can change. So right now, we’re optimizing for these last two."

Why won't Taste and Agency be equalized?

Taste is your value system—the ability to make decisions when there is no correct answer. When deciding which direction a product should take or how to balance trade-offs in an architecture, AI can offer suggestions, but determining which suggestion is right still requires someone with genuine judgment. Taste is rooted in aesthetics and values, and it cannot be significantly altered by effort alone in the short term.

Agency is the willpower to drive things forward—taking initiative without waiting for instructions, persisting through obstacles, and seeing half-finished tasks through to completion. This is something AI cannot provide.

Previously, hiring focused on experience; later, Silicon Valley shifted to valuing Slope (growth rate), replacing past accumulation with learning speed. But Ivan says even Slope is no longer enough—it still measures only the rate of acquiring Capability, essentially spinning within the same depreciating dimension. Taste and Agency exist on an entirely different axis; their presence cannot be predicted by how quickly one learns.

Two actions related to hiring

The engineering role is hiring a large number of recent graduates, valuing initiative, curiosity, and judgment more than past experience; for the sales role, the first-round interview has eliminated resumes, requiring candidates to create something first, focusing on what they can do now and whether they’re willing to take the initiative. Both actions are doing the same thing: replacing “What have you done in the past?” with “Who are you right now?”

Consider a few questions: What convinced you the last time you decided to hire someone? Was it that the candidate had done something similar at a previous company, did their resume include credentials you respected, or was the scale of their past projects sufficiently large?

These are all signals of Capability. Without a method to assess Taste and Agency, your hiring process is likely still optimizing a dimension that is losing value.

Three years ago, Notion internally adopted a motto: We want to be a Jazz Band, not a Marching Band.

The fundamental difference isn’t about speed, but about who can improvise. A march band needs a conductor; each musician follows the score, and uniformity is a virtue. A jazz band has structure and mutual understanding, but each member can at any moment pick up on others and drive the music forward improvisationally. The conductor disappears, but the structure doesn’t—it’s been internalized by everyone.

Ivan says this is his self-calibration mechanism. He’s the type of person who can’t stand the idea of delegating everything and merely giving orders. Once he has a clear vision, he systematically recruits like-minded individuals to build a company that reflects his own character.

This logic is manifested in three specific actions within the organization.

Dumbbell-shaped engineering team

The Notion engineering team currently has a dumbbell shape: the ends consist of Super Juniors and Super Seniors, with the middle layer shrinking.

Previously, the value of a Senior Engineer was multidimensional: writing more reliable code, possessing deeper system understanding, and independently driving complex projects. Since the emergence of AI coding agents, most of these value dimensions have been taken over. As a result, the Senior Engineer’s value is now refocused on what remains: architectural judgment and directional insight.

LLMs are still weak when it comes to system architecture; their individual suggestions may seem reasonable, but when combined in complex systems, they often lead to issues—this is where taste becomes essential, and where only a select few top-tier seniors are truly irreplaceable.

The optimal team structure Ivan described is roughly this: one top-tier senior architect leading two or three junior engineers, each of whom manages two or three Coding Agents. This structure yields higher output and better multiplicative effects compared to a model where multiple seniors each manage their own agents. The middle layer is being squeezed from both ends—the execution layer is taken over by juniors and agents, while only truly skilled senior architects can occupy the decision-making layer; the value of intermediate roles is becoming increasingly unclear.

Dissolve the CMO organization

Notion currently does not have a CMO. Marketing has been divided into two independent tracks: one aligned with the product, directly connected to social media, and synchronized with product release cycles; the other supporting sales, focused on lead and demand generation.

The reason for removing the intermediate coordination layer is simple: after AI takes on the bulk of information transmission and coordination, the overhead of routing information through the CMO and then redistributing it has become too costly. When both sides handle their own tasks directly, everything happens faster.

Invite dozens of entrepreneurs

Notion has brought in a large number of founders with entrepreneurial experience through acquisitions, each leading the areas they know best. The person in charge of the meeting notes feature previously founded a startup specializing in meeting notes; the person responsible for enterprise search was the founder of an enterprise search product. By providing them with a better platform and resources to continue doing what they do best, Notion inherently follows a retention strategy.

Ivan himself is also a "Refounder," able to step into any area at any time—or step away completely—without either side feeling territorial. This reinforces the organization’s jazz band nature at the personnel level: those who join are already capable of playing independently.

Ivan broke down the planning into two fundamentally different tasks, handling each with entirely different logic.

He believes the financial plan is still useful—like the speed setting on a treadmill: when you set it to a certain level, you know exactly what pace you're running at, and that reading is real. Notion takes a conservative to neutral approach to finances, leaving ample buffer. In the AI era, costs have become a new variable—token expenses scale directly with product usage and must be carefully accounted for.

Product strategy is a different matter.

There are no plans—really, none—not six months, not three months, it’s week-by-week improvisation.

This judgment comes directly from the lessons of the second reconstruction. At the end of 2022, Notion wanted to build an AI Agent product and pushed hard on it. After a year and a half, there was almost no progress—not because the team wasn’t working hard, but because the underlying models simply weren’t ready yet. Any product plan at that stage was meaningless; what truly mattered was improvising continuously within the limits of what the technology allowed.

The only thing you can plan is the tempo; financial goals define the treadmill’s speed. The melody is improvised, written week by week based on the actual state of technology and the market. This is precisely why a jazz band is better suited to the present than a marching band: a marching band must have its entire score rehearsed in advance, while a jazz band improvises on the spot, not knowing where the next measure will go—but capable of catching it in the moment.

Ivan was asked what the organization would look like in three to four years, and instead of describing any technical blueprint, he first asked: What won’t change?

His answer is human nature. Humans are inherently hierarchical; division of labor has meaning; people have different interests and values—these have been constants for thousands of years. Even in legal systems, there are no autonomous corporations; CEOs and CFOs still must sign and take responsibility. These constants serve as anchors for organizational design; AI changes only the way information and decisions flow between people, not human nature itself.

But above this anchor, three layers of rewriting are already taking place. It’s worth seriously asking yourself three questions:

• Is the hiring process still primarily focused on optimizing capability? Are there established methods to assess taste and agency?
• In your organization, how many people have core values centered on transmitting information and executing instructions? The structural pressures on these roles will continue to increase as AI tools mature.
• Are you still trying to schedule out a six-month roadmap for your product? It’s not that quarterly planning itself is flawed—it’s whether you’re treating it as a commitment or as a reference that’s adjusted weekly.

Finally:

Modern knowledge work is only about 150 years old. It was invented. It’s not as old as fire or language. Why can’t it be a new variation of it?

Knowledge work has only existed for 150 years—it was invented by humans, and the logic of how companies operate was also created by humans. If it can be invented, it can be rewritten. Notion is already rewriting it, and it did so two years earlier than most.