Notion CEO: AI companies should be a 'Jazz Band,' and I am a 'Refounder'

marsbitPublicado a 2026-05-26Actualizado a 2026-05-26

Resumen

Notion CEO Ivan Zhao, in a recent podcast, shared his journey of twice rebuilding the company from near-collapse and now applying the same "Refounder" mindset to reshape the 1000-person organization in the AI era. He argues that AI has commoditized technical capability (Capability). True talent now hinges on Taste (judgment/values) and Agency (proactive drive), necessitating a shift in hiring—e.g., hiring more juniors for curiosity and having sales candidates demonstrate work upfront. Zhao envisions the company as a "Jazz Band"—agile and improvisational—versus a rigid "Marching Band." This is reflected in an engineering "dumbbell" structure (super juniors + top-tier seniors, with middle layers compressed), dissolving the CMO role to let teams operate directly, and integrating entrepreneurs via acquisitions to lead their expertise areas. Notion has abandoned traditional long-term product roadmaps, planning only conservatively for finances while adopting a week-by-week, improvisational approach to product strategy, as longer plans proved futile during rapid AI shifts. He concludes that while human nature and roles remain constants, companies must rewrite their approaches to hiring (valuing Taste/Agency over Capability), organizational design (reducing roles focused on coordination/execution), and planning (embracing flexibility). Modern knowledge work, being only ~150 years old, is ripe for reinvention.

Last week, Notion CEO Ivan Zhao recorded a podcast at Sequoia Capital, discussing his experiences of twice bringing the company back from the brink of death and twice rebuilding from the ground up. Now, he is using the same logic to 're-found' this thousand-person company, calling himself a 'Refounder':

He believes AI has turned technical capability into a commodity, and what is truly scarce now is Taste and Agency. Therefore, hiring standards must change; information dissemination and coordination work are being taken over by AI, so organizational structure must change; technology changes too fast, making any planning beyond a few weeks likely obsolete, so planning methods must also change. Enjoy!

01. How Notion Was Rebuilt Twice from Near-Death Experiences

In 2015, after two years, Notion still hadn't found Product-Market Fit (PMF). Money was running out. Ivan and co-founder Simon made a decision most founders wouldn't dare: lay off everyone, move to Kyoto, and rebuild from scratch. They sublet their residences and office in San Francisco. During that time, Notion surprisingly achieved its first positive cash flow.

(Notion's initial office in San Francisco)

After arriving in Kyoto, life became extremely simple. Write code, eat, write code again, eat again. No team, no processes, no resources, just two people and an idea. This experience taught Ivan for the first time: what drives things forward has always relied on judgment and willpower; how abundant resources are is secondary. A year and a half later, Notion 1.0 launched.

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

The second time was in 2023. The team was having an offsite in Cancun, and Ivan got early access to GPT-4. That experience was an almost shocking impact for him. He instantly judged: this changes everything, and if we don't bet the entire company on it, nothing else we do will matter. So he announced a restart within the 500-person company, fully pivoting to AI.

But what followed was nearly a year and a half of agony. Model technology wasn't mature yet. They tried almost every direction, none of which worked. Growth stalled, morale plummeted. It wasn't until the underlying models truly matured that the product began to take off, with the revenue inflection point and the AI product's traction appearing almost simultaneously.

In both experiences, what truly mattered was his judgment and the willpower to keep moving forward amidst uncertainty, which also became the starting point for his later 're-founding' of the company.

02. 'Capability' Is Depreciating, Yet Companies Still Pay a Premium for It

Ivan proposed a talent formula:

Talent = Capability × Taste × Agency

Understanding this formula lies in the derivation process.

Why is Capability Depreciating?

Before Google, information access was a scarce resource, and those who could find information had a real competitive advantage. After Google appeared, this advantage vanished. 'I can find this information' became a basic skill. AI is causing the same thing to happen at the level of capability production. Writing code, crafting copy, performing data analysis—tasks that used to require years of accumulation to do well—can now be done to a decent level by recent graduates equipped with AI tools. The scarcity of Capability is being systematically compressed.

Ivan's original words were: 'What LLMs do is like how Google gave everyone access to information; they give everyone the ability to be a decent writer and programmer. Everyone possesses Capability. But Taste still matters—it's your value system, it's your manifestation of what you want to bring to the world; Agency is the same—how hard you work, this is something companies cannot change. So now we're optimizing for the latter two.'

Why Won't Taste and Agency Be Leveled?

Taste is your value system, the ability to make judgments when there's no standard answer. Which direction a product should take, how an architecture should be balanced—AI can give suggestions, but deciding which suggestion is right still requires someone with real judgment to make the call. Taste is rooted in aesthetics and values, something that can't be changed much by effort in the short term.

Agency is the willpower to make things happen. Moving forward without waiting for instructions, not backing down in the face of resistance, truly completing a half-finished task. This is also something AI cannot provide.

Previously, hiring looked at experience; later, Silicon Valley trended towards looking at Slope (growth rate), using learning speed to replace past accumulation. But Ivan says even Slope isn't enough now—it still measures the acquisition speed of Capability, essentially still operating within the same depreciating dimension. Taste and Agency are on entirely different axes; how fast one learns cannot predict them.

Two Actions in Hiring

For engineering roles, they hire many recent graduates, focusing not on past experience but on initiative, curiosity, and judgment; for sales roles, the first round of interviews eliminates resumes, requiring candidates to first build something, focusing on what they can do now and their willingness to take initiative. Both actions are doing the same thing: replacing 'what you have done' with 'what kind of person you are now.'

Consider these questions: What was the last reason that convinced you to hire someone? Was it that the candidate had done similar work at a certain company, had a background you respected on their resume, or that past project scales were large enough?

These are all signals of Capability. Without a set of methods to evaluate Taste and Agency, your hiring process is likely still optimizing a depreciating dimension.

03. Becoming an Agile Jazz Band

Three years ago, an internal slogan was established at Notion: We want to be a Jazz Band, not a Marching Band.

The fundamental difference isn't in tempo, but in who can improvise. A marching band needs a conductor; each musician follows the score, and uniformity is a virtue. A jazz band has structure and默契 (tacit understanding), but everyone can pick up from others and improvise at any moment. The conductor disappears, but structure doesn't, because it's already internalized within each person.

Ivan says this is his self-calibration mechanism. He is a jazz-band type of person and cannot stand the feeling of delegating everything and just giving orders. Once he figured it out, he systematically began recruiting like-minded people to build a company aligned with his own temperament.

This logic manifests in the organization through three specific actions.

The 'Dumbbell-Shaped' Engineering Team

The shape of Notion's engineering team is now a dumbbell: the two ends are Super Juniors and Super Seniors, with the middle layer shrinking.

Previously, the value of a Senior Engineer was multi-dimensional: more reliable code, deeper system understanding, ability to independently drive complex projects. After the emergence of AI Coding Agents, most of this value sequence began to be taken over. Therefore, the value of Seniors has been refocused on the remaining parts: architectural judgment and directional sense.

LLMs are still quite weak at system architecture; suggestions that look reasonable individually often cause problems when pieced together in complex systems. This is where Taste is required, the truly irreplaceable domain of a few top-tier Seniors.

Ivan described the optimal combination as something like this: a top-tier senior architect leads two or three young engineers, each managing two or three Coding Agents. Compared to a group of Seniors each managing Agents, this structure yields higher output and better multiplier effects. The middle layer is compressed from both ends; the execution layer is taken over by Juniors + Agents; the judgment layer is only held by top Seniors with genuine architectural ability. The value of intermediate positions is becoming increasingly unclear.

Dissolving the CMO Organization

Notion currently has no CMO. Marketing has been split into two independently operating lines: one close to the product, directly connected to social media, following the product release rhythm; the other serving sales, focusing on lead and demand generation.

The reason for removing the middle coordination layer is simple: after AI takes over a large amount of information transmission and coordination work, the cost of letting information make a detour through the CMO before distribution is now too expensive. Each side handling its own affairs is faster.

Introducing Dozens of Entrepreneurs

Notion has brought in many founders with entrepreneurial experience through acquisitions, each leading the domain they know best. The person responsible for the meeting notes feature previously ran a startup focused on meeting notes; the person responsible for enterprise search was the founder of an enterprise search product. Giving them a better platform and resources to continue doing what they excel at is itself a logic for retention.

Ivan himself is also a 'Refounder,' able to jump into any area at any time or fully let go; there's no territorial threat on either side. This reinforces the jazz band property of the organization at the personnel composition level—bringing in people who can already play independently.

04. Notion Has Given Up on Product Planning

Ivan splits planning into two fundamentally different things, handled with completely different logic.

He believes financial planning is still useful, like the speed setting on a treadmill. You set it to a certain level, and you know what pace you're running at; this reading is real. Notion tends to be conservative to neutral in finance, leaving itself enough Buffer. In the AI era, cost has also become a new variable; Token expenses can scale directly with product usage and must be seriously factored in.

Product strategy is another matter entirely.

No plans, really none—not six months, not three months, but improvising week by week.

This judgment comes directly from the lessons of the second rebuild. At the end of 2022, Notion wanted to build an AI Agent product and was pushing it very seriously. There was almost no progress for a year and a half. The team wasn't slacking; it was because the underlying models themselves weren't ready yet. Any product plan at that stage was empty. What truly worked was constant improvisation within the boundaries set by technology.

You can only plan the Tempo; financial goals define the treadmill speed. The Melody is improvised, written weekly based on the actual conditions of technology and the market. This is precisely the core reason why a jazz band is more suitable for the present than a marching band: a marching band must arrange all the sheet music in advance before going on stage; a jazz band improvises and adapts on the spot, not knowing where the next bar will go but having the ability to catch it in the moment.

05. At Which Level Has Your Company Not Yet Started Rewriting?

When asked what the organization would look like three or four years later, Ivan didn't describe any technological blueprint but first asked: What remains unchanged?

His answer is human nature. Humans are inherently hierarchical; division of labor has meaning; people have different interests and values—these are constants over millennia. The legal system also has no autonomous companies; CEOs and CFOs still have to sign and take responsibility. These invariants are the anchor points for organizational design. What AI changes is the way information and decision flows between these people; human nature itself cannot be moved.

But above this anchor point, rewriting is already happening at three levels. It's worth seriously asking yourself three questions:

  • Is your hiring process still primarily optimizing for Capability? Do you have a method for evaluating Taste, a method for evaluating Agency?
  • In your organization, how many people's core value lies in transmitting information and executing instructions? The structural pressure on these positions will continue to increase as AI tools mature.
  • Are your product plans still trying to pre-arrange a six-month score? This isn't to say quarterly planning itself is problematic, but rather, how you use it—as a promise, or as a reference adjusted 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 only existed for about 150 years; it's a human invention. The logic of how companies run is also man-made. What can be invented can be rewritten. Notion is already rewriting it, and it started two years earlier than most.

Preguntas relacionadas

QAccording to Ivan Zhao, what are the two key human qualities that remain valuable in the age of AI, and why?

AAccording to Ivan Zhao, the two key human qualities are Taste and Agency. Taste, which is one's value system and ability to make judgments without clear answers, remains crucial because AI can offer suggestions but cannot determine which suggestion is correct. Agency, the willpower to drive things forward, is equally important because it involves taking initiative and persevering through challenges—traits that AI cannot replicate.

QWhat are the three specific organizational changes Notion implemented to become a 'Jazz Band'?

AThe three specific organizational changes are: 1) Creating a 'dumbbell-shaped' engineering team consisting of Super Juniors and Super Seniors, with the middle layer shrinking. 2) Dissolving the CMO organization and splitting marketing into two independent lines: one for product-related social media and another for sales support. 3) Introducing dozens of entrepreneurs through acquisitions to lead areas they are passionate about, enhancing the organization's ability to improvise and adapt.

QHow did Notion's approach to product planning change after Ivan Zhao's second 'rebuilding' experience?

AAfter the second rebuilding experience, Notion abandoned traditional long-term product planning. Instead, they adopted a weekly improvisation approach, where product strategy is adjusted weekly based on real-time technological and market conditions. Financial planning remains important for setting the pace, but product direction is treated like a melody in a jazz performance—improvised and responsive to the moment.

QWhat does Ivan Zhao mean by 'Refounder,' and how does this concept apply to his leadership at Notion?

AThe term 'Refounder' refers to Ivan Zhao's approach of repeatedly rebuilding and reinventing Notion from the ground up. It applies to his leadership by emphasizing adaptability, judgment, and willpower in navigating uncertainty. As a Refounder, he can dive into any area of the company or step back entirely, fostering a culture where flexibility and continuous reinvention are central to the organization's strategy.

QIn Ivan Zhao's view, what is the fundamental difference between a 'Jazz Band' and a 'Marching Band' in an organizational context?

AIn an organizational context, a 'Marching Band' operates under a centralized command structure where each member follows a pre-set plan. In contrast, a 'Jazz Band' relies on shared structure and默契, allowing members to improvise, adapt, and take initiative independently. The key difference lies in flexibility: a Jazz Band can respond dynamically to changes without relying on top-down指令, making it better suited for fast-paced, uncertain environments.

Lecturas Relacionadas

Del Código a la Cognición: Una Guía de Diez Mil Palabras sobre la Evolución del Cerebro Robótico

Desde el código clásico hasta los modelos que simulan la realidad: así ha evolucionado la inteligencia de los robots. Durante décadas, dependieron de software programado manualmente (percepción, planificación, control) para tareas específicas pero rígidas. La llegada del *deep learning* y el aprendizaje por refuerzo mejoró la percepción y el control, aunque con escasa capacidad de generalización. La aparición de los grandes modelos de lenguaje (LLM) marcó un punto de inflexión, actuando como "compiladores de lenguaje natural" que traducen órdenes en planes de acción ejecutables por sistemas como ROS. Sin embargo, el avance definitivo llegó con los **Modelos Visión-Lenguaje-Acción (VLA)**, que fusionan la percepción visual, la comprensión del lenguaje y la generación de movimientos en una sola red neuronal, permitiendo una adaptación mucho mayor. Los robots más avanzados actualmente utilizan una **arquitectura de doble cerebro**: un sistema lento (S2) para el razonamiento de alto nivel y uno rápido (S1) para el control motor reactivo, a veces con un tercer nivel reflejo (S0) para el equilibrio. Todo el procesamiento crítico se ejecuta localmente en el robot por motivos de seguridad y latencia. El siguiente gran salto son los **Modelos del Mundo**. Estas redes no predicen la siguiente acción, sino las consecuencias físicas de las acciones posibles, permitiendo al robot "imaginar" y evaluar futuros antes de actuar. Esto mejora drásticamente la recuperación ante errores, la planificación a largo plazo y la generalización. Modelos como NVIDIA Cosmos, DeepMind Genie y Meta V-JEPA exploran diferentes enfoques arquitectónicos para lograrlo. Aunque los precios del hardware caen y los modelos open-source aceleran el desarrollo, desafíos como la eficiencia en el aprendizaje, la generalización entre robots físicos y el razonamiento de sentido común persisten. La inteligencia robótica está en plena evolución, transitando desde un código predecible hacia sistemas capaces de aprender y simular el mundo que los rodea.

marsbitHace 19 min(s)

Del Código a la Cognición: Una Guía de Diez Mil Palabras sobre la Evolución del Cerebro Robótico

marsbitHace 19 min(s)

La burbuja de la IA se está pinchando

En los últimos días, la volatilidad del mercado ha avivado el debate sobre la "burbuja de la IA". Figuras como Ray Dalio advierten sobre niveles de burbuja "relativamente altos", mientras que Jensen Huang de NVIDIA ve una oportunidad enorme, con la demanda de potencia de cálculo apenas comenzando. Ambos tienen razón. La industria de la IA presenta inevitablemente una burbuja, un fenómeno común al surgir una fuerza productiva disruptiva, similar a la burbuja de las puntocom. Aunque aquella crisis evaporó billones, sentó la infraestructura física (fibra óptica, redes) que luego permitió el auge de gigantes como Amazon, Netflix o la nube. Se subestimó su impacto a largo plazo. Hoy, la inversión es masiva: se prevé que los principales proveedores de nube gasten 690.000 millones de dólares en 2026 en infraestructura física (enfriamiento, energía, redes), muy por encima de los ingresos actuales de las empresas de IA pura. Sin embargo, los costes se han desplomado: el precio por millón de tokens ha caído más de un 99.7% desde 2023. Según la "Paradoja de Jevons", esta eficiencia no reduce el consumo, sino que lo dispara al desbloquear nuevas aplicaciones (agentes autónomos, RAG, multimodalidad). Empresas de todos los sectores (finanzas, salud, manufactura) ya adoptan la "IA+", enfocándose en su implementación, no en si usarla. La burbuja se está desinflando, eliminando start-ups sin sustento, lo que purifica el mercado. El valor migrará gradualmente de la infraestructura (CapEx) a las aplicaciones que optimicen operaciones (OpEx). Aunque las valoraciones son elevadas, un crecimiento sólido de los ingresos podría digerirlas con el tiempo. Estamos en un punto de inflexión. Tras el inevitable ajuste, la infraestructura y los algoritmos optimizados, ahora baratos, impulsarán una nueva era en la que la IA integrará y potenciará todas las industrias. El ruido de la burbuja pasará; el avance de la fuerza productiva subyacente permanece.

链捕手Hace 33 min(s)

La burbuja de la IA se está pinchando

链捕手Hace 33 min(s)

La burbuja de la IA se está desinflando

El burbuja de la IA está estallando, pero esto es parte del proceso natural de una nueva tecnología disruptiva. Aunque existen paralelismos con la burbuja puntocom del 2000, donde se produjo una corrección severa, la infraestructura física construida durante ese período allanó el camino para gigantes posteriores. Actualmente, las grandes tecnológicas están invirtiendo masivamente en infraestructura para la IA, mientras que los ingresos de las empresas de IA pura son mucho menores, lo que sugiere una asimetría. Sin embargo, un factor clave es el drástico descenso de más del 99.7% en el coste de la inferencia de IA, lo que, paradójicamente, ha impulsado un aumento exponencial en su uso empresarial (Paradoja de Jevons). Las empresas ahora implementan agentes de IA para tareas complejas. El mercado se está depurando, eliminando empresas con solo ideas y sin propuestas de valor sólidas. El valor se está desplazando gradualmente de los proveedores de infraestructura (CapEx) hacia las aplicaciones que optimizan operaciones (OpEx). A pesar del ajuste en los mercados de capitales, la adopción de la IA en sectores como la fabricación, las finanzas, la medicina y el derecho es profunda e irreversible. La corrección actual elimina la especulación, pero la tendencia subyacente es clara: al igual que Internet, la IA se integrará en todas las industrias, impulsando una nueva era de productividad.

marsbitHace 34 min(s)

La burbuja de la IA se está desinflando

marsbitHace 34 min(s)

El Próximo Boom de los ETF de Bitcoin Podría Venir de Japón: He Aquí Por Qué

Los ETF de Bitcoin al contado en EE.UU. atraviesan un rendimiento bajista persistente, con salidas netas durante 13 días consecutivos entre mediados de mayo y principios de junio, totalizando unos 4.330 millones de dólares retirados. A pesar de ello, estos activos mantienen un valor neto de 75.120 millones de dólares. Mientras, los analistas comienzan a anticipar qué país podría albergar el próximo gran mercado de ETF, destacando Japón como candidato clave. Las reformas regulatorias en Japón, que trasladarían las criptoactivos al marco de la Ley de Instrumentos Financieros, están fortaleciendo las perspectivas para un ETF de Bitcoin. Se estima que, en un escenario conservador, un ETF japonés podría atraer hasta 900.000 millones de yenes (5.610 millones de dólares), dados los aproximadamente 2.350 billones de yenes en activos financieros de los hogares. En un caso base, los depósitos podrían alcanzar 1,4 billones de yenes (8.730 millones de dólares) tras el lanzamiento, representando una demanda de unas 140.000 BTC al precio actual. Un escenario alcista podría ver entradas de hasta 3,1 billones de yenes (19.340 millones de dólares) en el primer año. Más allá de la apreciación del precio, un ETF al contado facilitaría la participación de inversores, asesores e instituciones, otorgando mayor legitimidad al Bitcoin en las finanzas tradicionales. Mientras, el precio de BTC ronda los 61.038 dólares, con un descenso del 2,81% en las últimas 24 horas.

bitcoinistHace 1 hora(s)

El Próximo Boom de los ETF de Bitcoin Podría Venir de Japón: He Aquí Por Qué

bitcoinistHace 1 hora(s)

Los 43 minutos de Trump: la narrativa del hombre fuerte se descontrola y la guerra mediática se intensifica

**Resumen del artículo en español:** Tras más de una semana sin aparecer en público, el presidente Donald Trump celebró una conferencia de prensa de 43 minutos que se alejó de los temas centrales (su salud, la acción militar en Irán, las fracturas en su partido) y derivó en un espectáculo político errático y lleno de quejas. La intervención comenzó hablando extensamente sobre la renovación de un estanque reflejante, comparó el tamaño de sus mítines con los de Martin Luther King, atacó a ciudades estadounidenses y a sus adversarios políticos, y culminó con un ataque personal a una periodista de CNN, a quien criticó por "nunca sonreír" y tener "odio en los ojos". El artículo subraya dos dimensiones clave. Primero, revela un estado presidencial que parece fuera de control, defensivo y volátil, evidenciado por el abrupto final del acto y el rápido desalojo del lugar por su equipo. Segundo, analiza cambios institucionales: Trump firmó una orden ejecutiva que elimina protecciones laborales para miles de altos funcionarios federales, priorizando la lealtad política sobre la experiencia y la independencia. El texto amplía el análisis al panorama mediático, señalando cómo la presión política y los intereses comerciales están erosionando la independencia periodística, citando el caso de la destitución de un veterano periodista en la CBS. Argumenta que, ante la captura de los grandes medios, el periodismo independiente se vuelve crucial para mantener los hechos. El autor hace un llamado a apoyar económicamente a las voces independientes como forma de resistencia, enfatizando que el ataque de Trump a la prensa es un ataque al derecho del público a estar informado. El artículo concluye notando que, el mismo día de este episodio, la Cámara de Representantes (con el voto a favor de cuatro republicanos) aprobó una resolución para poner fin a la guerra en Irán, mostrando grietas en su partido y sugiriendo que su comportamiento extremo está empezando a alienar a sus propios aliados.

marsbitHace 5 hora(s)

Los 43 minutos de Trump: la narrativa del hombre fuerte se descontrola y la guerra mediática se intensifica

marsbitHace 5 hora(s)

Trading

Spot
Futuros

Artículos destacados

Cómo comprar BAND

¡Bienvenido a HTX.com! Hemos hecho que comprar Band Protocol (BAND) sea simple y conveniente. Sigue nuestra guía paso a paso para iniciar tu viaje de criptos.Paso 1: crea tu cuenta HTXUtiliza tu correo electrónico o número de teléfono para registrarte y obtener una cuenta gratuita en HTX. Experimenta un proceso de registro sin complicaciones y desbloquea todas las funciones.Obtener mi cuentaPaso 2: ve a Comprar cripto y elige tu método de pagoTarjeta de crédito/débito: usa tu Visa o Mastercard para comprar Band Protocol (BAND) al instante.Saldo: utiliza fondos del saldo de tu cuenta HTX para tradear sin problemas.Terceros: hemos agregado métodos de pago populares como Google Pay y Apple Pay para mejorar la comodidad.P2P: tradear directamente con otros usuarios en HTX.Over-the-Counter (OTC): ofrecemos servicios personalizados y tipos de cambio competitivos para los traders.Paso 3: guarda tu Band Protocol (BAND)Después de comprar tu Band Protocol (BAND), guárdalo en tu cuenta HTX. Alternativamente, puedes enviarlo a otro lugar mediante transferencia blockchain o utilizarlo para tradear otras criptomonedas.Paso 4: tradear Band Protocol (BAND)Tradear fácilmente con Band Protocol (BAND) en HTX's mercado spot. Simplemente accede a tu cuenta, selecciona tu par de trading, ejecuta tus trades y monitorea en tiempo real. Ofrecemos una experiencia fácil de usar tanto para principiantes como para traders experimentados.

120 Vistas totalesPublicado en 2024.12.12Actualizado en 2026.06.02

Cómo comprar BAND

Discusiones

Bienvenido a la comunidad de HTX. Aquí puedes mantenerte informado sobre los últimos desarrollos de la plataforma y acceder a análisis profesionales del mercado. A continuación se presentan las opiniones de los usuarios sobre el precio de BAND (BAND).

活动图片