The Refounder — Notion's Ivan Zhao's playbook for the AI era

In the new episode of Sequoia Capital’s Training Data podcast, Notion’s founder Ivan Zhao isn’t called a founder anymore. He’s called a refounder — a founder who restarts his own company. Zhao has done it twice: first in 2015 in Kyoto when Notion was stuck pre-PMF, and again in 2023 at an offsite in Cancun when generative AI rewrote the rules. That’s the heart of the episode, and it leaves a useful question for the Finnish growth founder: does your company need a refounder moment now?

Two restarts — and why it’s the new normal

Notion in 2015 was a classic pre-PMF story: money burning, direction lost, two cofounders looking for a reason to keep going. In Kyoto, Zhao and his cofounder sat down and rebuilt the company from the studs. Only after that did Notion find product–market fit, and only after that did the story we all know begin. The company stayed small for years (under ten people still in 2019) and became profitable precisely because of its smallness.

The second refounder moment came in 2023. Generative AI changed in a single year what a SaaS company is actually for. Zhao took the team to Cancun for an offsite and rebuilt the company once more — this time as AI-native. Sequoia’s Pat Grady calls Zhao “the canonical example of how a SaaS company becomes an AI company.” It didn’t mean a new product layered on the old one. It meant a new way to organise work, hire people and reward them.

For Finnish founders this is the most important sentence of the year so far: if you’re building a software company that hasn’t yet had its refounder moment around AI, you are already late. Not so late the game is lost — but late enough that next year is the last sensible time to do it.

“Jazz band, not a marching band” — Zhao’s mantra

The episode’s through-line is Zhao’s company mantra: we are not a marching band, we are a jazz band. The marching band plays from sheet music, step by step, conductor-led. The jazz band improvises, listens to itself, and produces something no individual player planned. “We’ve been hiring more and more jazz band people,” Zhao says. “And AI has made them stars.”

Concretely this means three things. First, role boundaries blur: Notion’s best development happens when designers, engineers and product managers sit around the same table, run their own evals and experiment. Not “PM talks to customers, designer draws, devs build”. No.

Second, building on top of an LLM is more like brewing beer than building a bridge. In classical software you can design and execute predictably. With a language model you have to throw your best people in and see what comes out. This runs against the Finnish engineering culture — and that’s why it matters so much.

Third: late in the episode, the Sequoia host floats a name for the new way of working, the one that comes after “manager mode” and “founder mode”. Zhao’s suggestion: Jasmo. Jazz mode.

AI is the steel for organisations

Zhao has one line worth reading twice. “Before steel, buildings could not be taller than five or six floors. New York was a six-floor city of iron and stone. After steel, everything changed. Language model plus software is the steel for organisations.”

What does that mean in practice? That the org chart, which has been the same for thousands of years (a triangle where information moves slowly top to bottom), can be rebuilt. Jack Dorsey is experimenting at Block with three-layer organisations where AI sits in the middle and people surround it. Brian Armstrong talks about Coinbase with five layers, everyone a player-coach. Zhao himself runs Notion with 7 direct reports, but his best managers already run 15–20.

The classic Finnish growth company span of control is 5–8. If Zhao’s thesis holds, the right number is closer to 15 — and the gap doesn’t close by hiring middle managers but by building AI into the middle of the organisation. This is so radical that most Finnish hiring decisions today are made for the wrong joint.

Different-shape hires — agency, not years

Two years ago Zhao began hiring “different shape engineers” for Notion. That means no longer optimising hires for years of experience, but for agency, energy and optimism. Who takes responsibility, who tries things, who believes tomorrow can be better. These are the lines from late in the episode worth putting on a wall.

The Finnish hiring tradition values “steady experience” and “track record at big houses”. At the refounder moment that isn’t an advantage — it’s a brake. A classic SaaS executive may be excellent operationally but isn’t necessarily the person you want sitting next to you in Cancun rewriting the product fifty.

What a Finnish founder can do tomorrow

Three practical steps you can take within a month.

1. Ask yourself honestly: does your company need a refounder moment? If the answer is “not yet”, book a three-day offsite at a hotel in some small town (Kyoto is taken, try Porvoo) and sit with it alone or with one other person. If the answer is “maybe”, keep it a two-person conversation before telling the team. If the answer is “yes”, start next Monday.




2. Decide where in your organisation AI sits in the middle, not on the side. Don’t ask “where could we use AI”. Ask: “where in our org does information move between people, and could it move through a language model faster?” The answer is usually the sales–product interface or operational reporting.




3. Flatten one layer and grow span of control with your best managers. If your managers currently have five reports, try seven with the best of them. See what breaks and fix it with AI. Then nine.

FAQ

Is “refounder” only for software companies?

No. The same logic applies to any knowledge-work business that builds services or processes. Consulting, media, financial firms, even municipal organisations can benefit from the refounder mindset. The concrete claim (language model + software = steel) works wherever information moves.

How do you know you’re at a refounder moment?

Three classic signs: (1) your product’s core assumptions were built for a world that no longer exists, (2) your best people are asking questions your current structure cannot answer, (3) your competitor is doing something you cannot explain with classical product development. If at least two apply, it’s time.

Won’t the Jasmo style just collapse once you’re big enough?

Zhao has thought about this. He doesn’t claim hierarchy disappears — “human nature is hierarchical” is his line. But the hierarchy can be much flatter than before, if AI sits in between. Not four-tier but two- or three-tier. Not five reports but fifteen.

What has Notion done differently in the past year?

Based on the Sequoia episode: changed the hiring criterion from experience to agency, grown span of control among best managers to 15+, moved product–design–engineering roles to the same table with shared evals, and built the org around the assumption that the LLM handles the “information-moving” middle work.

Watch the full episode

Sequoia’s Training Data podcast episode with Zhao runs over an hour and is one of the year’s most important startup interviews. Short clips don’t do it justice — watch the whole thing below.

▶

Source: Sequoia Capital — Training Data podcast, 21 May 2026