Notion CEO: AI Companies Should Be “Jazz Bands,” and I Am a “Refounder”
TechFlow Selected TechFlow Selected
Notion CEO: AI Companies Should Be “Jazz Bands,” and I Am a “Refounder”
Information transmission and coordination are being taken over by AI, so organizational structures must change.
Navigating Web3 tides with focused insightsLinkloud Introduction
Last week, Notion CEO Ivan Zhao recorded a podcast with Sequoia Capital, recounting his experiences of steering the company through two near-death crises—and rebuilding it from scratch twice. Today, he’s applying the same logic to “reconstruct” this 1,000-person company—and calls himself a “Refounder”:
He believes AI has commoditized technical capability, making Taste (aesthetic judgment) and Agency (initiative) the truly scarce assets—so hiring criteria must change; AI is increasingly handling information dissemination and coordination, so organizational structure must change; and with technology evolving so rapidly, any plan longer than a few weeks risks obsolescence—so planning methodology itself must change. Enjoy!
01. How Notion Rebuilt Itself Twice From the Brink
In 2015, Notion spent two years failing to achieve product-market fit (PMF). Cash was running dangerously low. Ivan and co-founder Simon made a decision most founders wouldn’t dare: they laid off the entire team and relocated to Kyoto to rebuild from zero. They sublet their San Francisco apartment and office—and during that period, Notion achieved positive cash flow for the first time.
(Notion’s original office in San Francisco)
Once settled, life became radically simple: code, eat, code again, eat again. No team, no processes, no resources—just two people and an idea. This experience gave Ivan his first real insight: what actually drives progress is judgment and willpower—not resource abundance. A year and a half later, Notion 1.0 launched.
(Ivan and co-founder Simon in their Kyoto apartment)
The second crisis came in 2023. While holding an offsite in Cancún, Ivan gained early access to GPT-4. That experience struck him with near-shock—he realized instantly: “This changes everything. If we don’t bet the entire company on it, nothing else matters.” He announced a full reset across the 500-person organization and pivoted aggressively toward AI.
What followed was nearly 18 months of intense struggle. Underlying models weren’t mature yet; they tried nearly every possible direction—and none worked. Growth stalled; morale plummeted. Only when foundational models truly matured did the product begin to take off—revenue inflection and AI product traction emerged almost simultaneously.
In both episodes, what truly mattered was Ivan’s judgment and his unwavering will to move forward amid uncertainty—this became the foundation for his later “rewriting” of the company.
02. “Capability” Is Depreciating—Yet Companies Still Pay a Premium for It
Ivan proposed a talent formula:
Talent = Capability × Taste × Agency (Talent = Capability × Judgment × Initiative).
Understanding this formula hinges on its derivation.
Why Is Capability Depreciating?
Before Google, access to information was a scarce resource—those who could find it held genuine competitive advantage. After Google launched, that advantage vanished; “I can find this information” became table stakes. AI is doing the same thing—but now at the level of capability production. Writing code, drafting copy, performing data analysis—tasks once requiring years of accumulated expertise—can now be done at a solid level by fresh graduates using AI tools. The scarcity of Capability is being systematically compressed.
As Ivan put it: “What LLMs do is like what Google did for information access—they make everyone a decent writer and programmer. Everyone now possesses Capability. But Taste remains critical: it reflects your value system and what you aim to bring to the world. Agency matters just as much: how hard you push forward is something no company can engineer for you. So today, we’re optimizing precisely those latter two dimensions.”
Why Can’t Taste and Agency Be Flattened?
Taste is your value system—the ability to make judgments where there are no clear right answers. Which direction should a product take? How should an architecture be designed? AI can offer suggestions—but choosing which suggestion is correct still requires real human judgment. Taste is rooted in aesthetics and values—effort alone rarely reshapes it significantly in the short term.
Agency is the willpower to drive things forward. It means acting without waiting for instructions, persisting despite obstacles, and seeing a half-finished task through to completion. This, too, is something AI cannot supply.
Historically, hiring prioritized experience. Later, Silicon Valley shifted toward evaluating Slope (learning trajectory)—using learning speed to proxy for past accumulation. But Ivan argues even Slope isn’t enough anymore: it still measures only the rate of Capability acquisition—operating within a dimension that’s actively depreciating. Taste and Agency exist on entirely different axes—speed of learning tells us nothing about them.
Two Concrete Hiring Actions
For engineering roles, Notion hires heavily from new grads—not for prior experience, but for initiative, curiosity, and judgment. For sales roles, the first-round interview drops resumes entirely: candidates must first build something tangible—revealing what they can do *now*, and whether they’ll act proactively. Both actions serve the same goal: replacing “What have you done?” with “Who are you *right now*?”
Consider a few questions: What convinced you to hire someone last time? Was it because they’d done similar work at another company? Because their resume featured a background you respected? Or because their past projects were large in scale?
All of these are signals of Capability. Without systematic methods to evaluate Taste and Agency, your hiring process is likely still optimizing a depreciating dimension.
03. Building a Flexible Jazz Band
Three years ago, Notion adopted an internal slogan: “We want to become a Jazz Band—not a Marching Band.”
The fundamental distinction isn’t tempo—it’s improvisation. A marching band needs a conductor; each musician follows sheet music precisely—uniformity is virtue. A jazz band operates within structure and shared understanding, yet any member can spontaneously pick up where another leaves off and push things forward. The conductor vanishes—but structure remains, internalized in every individual.
Ivan describes this as his personal calibration mechanism. He’s a “jazz-band person”—he can’t tolerate delegating everything and merely issuing commands. Once he clarified this, he began systematically recruiting others like himself, building a company aligned with his own temperament.
This logic manifests in three concrete organizational actions.
A “Dumbbell-Shaped” Engineering Team
Notion’s engineering team today resembles a dumbbell: heavy at both ends—Super Juniors (exceptionally junior engineers) and Super Seniors (exceptionally senior engineers)—with the middle layer shrinking.
Traditionally, Senior Engineers delivered multidimensional value: more reliable code, deeper system understanding, autonomy in driving complex projects. With AI Coding Agents emerging, much of that value has been absorbed. Senior value is thus refocused on what remains: architectural judgment and strategic direction.
LLMs remain weak at system architecture—individual suggestions may seem reasonable, but when stitched together into complex systems, they often fail. That’s where Taste comes in: the rare top-tier Senior’s irreplaceable edge.
Ivan describes the optimal configuration as follows: one elite senior architect leading two or three junior engineers, each managing two or three coding agents. This structure yields higher output and stronger multiplicative effects than a group of seniors each overseeing their own agents. The middle layer is squeezed from both ends: execution is handled by Juniors + Agents; judgment resides solely with elite architects possessing genuine architectural skill—the value of the “middle” position grows increasingly ambiguous.
Dissolving the CMO Function
Notion no longer has a CMO. Marketing is split into two independent, parallel tracks: one tightly integrated with product development and directly tied to social media, moving in lockstep with product releases; the other dedicated to sales enablement, focused on lead generation and demand creation.
The reason for eliminating the centralized coordination layer is simple: once AI handles much of information transfer and coordination, routing messages through a CMO before redistribution incurs excessive overhead. Letting each track manage its own domain independently proves faster.
Bringing in Dozens of Founders
Notion has acquired numerous startups, integrating their founders—each taking ownership of the domain they know best. The person leading meeting-note functionality previously ran a startup specializing in meeting notes; the person heading enterprise search was formerly the founder of an enterprise search product. Giving them better platforms and resources to continue excelling at what they do best is itself a retention strategy.
Ivan himself is a “Refounder”: he can dive deep into any domain—or step back entirely. Neither side perceives territorial threat. This strengthens the organization’s “jazz band” character at the personnel level: everyone joining is already capable of independent performance.
04. Notion Has Abandoned Product Planning Altogether
Ivan separates “planning” into two fundamentally distinct activities—and treats them with entirely different logics.
Financial planning, he believes, remains useful—like setting treadmill speed: you choose a pace and know your current running velocity. That metric is real. Notion adopts a conservative-to-neutral financial stance, building ample buffer. In the AI era, cost is also a new variable: token consumption scales directly with product usage and must be rigorously modeled.
Product strategy is another matter entirely.
There is no plan—not six months out, not three months out—just weekly improvisation.
This conclusion stems directly from lessons learned during the second rebuild. At the end of 2022, Notion earnestly pursued an AI Agent product—but made almost no progress over 18 months. The team wasn’t underperforming; the underlying models simply weren’t ready. Any product plan at that stage would have been hollow. What truly worked was continuous improvisation—within the constraints set by available technology.
You can only plan Tempo (rhythm): financial goals define treadmill speed. Melody (melody) is improvised—written anew each week based on actual technological and market conditions. This is precisely why the jazz band model fits today’s reality better than the marching band: the latter must rehearse every note in advance; the former improvises on the spot—no one knows where the next bar will go, but everyone can respond meaningfully in real time.
05. At Which Level Has Your Company Not Yet Begun Rewriting?
When asked what Notion’s organization might look like three or four years out, Ivan didn’t sketch a technical blueprint. Instead, he asked: “What won’t change?”
His answer: human nature. People are inherently hierarchical; division of labor is meaningful; individuals hold diverse interests and values—these are millennia-old constants. Legal frameworks don’t recognize self-governing corporations; CEOs and CFOs still bear signature-level responsibility. These invariants anchor organizational design. AI transforms *how* people exchange information and route decisions—but human nature itself remains untouched.
Yet atop this anchor, rewriting is already underway across three levels. Ask yourself three critical questions:
- Does your hiring process still primarily optimize for Capability? Do you have robust methods to assess Taste? To assess Agency?
- How many people in your organization derive core value from transmitting information and executing instructions? These roles face intensifying structural pressure as AI tools mature.
- Does your product planning still attempt to pre-compose a six-month score? This isn’t about quarterly planning per se—but whether you treat it as a binding commitment, or as a flexible reference point revised weekly?
Finally:
"Modern knowledge work is only about 150 years old. It's invented. It's not as old as fire or language. Why can't it be a new flavor of it？"
Knowledge work has existed for only 150 years—it’s a human invention. So is the operating logic of companies. What’s invented can be rewritten. Notion is already rewriting—and it started two years ahead of most.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
