Leadership Insights

What Really Happened to the Spotify Model: Hosanna Ouellette and Nora Bereczkei on autonomy, agents, and the 26 rocket ships

Episode notes

The conversation, captured

Date: August 4, 2026
Participants: Kristian Luoma Hosanna Ouellette Nora Bereczkei

The Coordination Tax is a podcast by In Parallel about what actually breaks down when teams stop sharing the same reality. In this episode, host Kristian Luoma sits down with Hosanna Ouellette and Nora Bereczkei, who together led the rebuild of Spotify’s product operating system. Nora has built the delivery backbone inside Spotify, the BBC and now at Gusto. Hosanna spent her career moving ideas through one of the largest matrix firms on the planet, and today deploys autonomous agents into live commercial workflows. They lived through what the coordination tax actually costs when hundreds of squads try to ship one unified experience to millions of users — and why a 10× productivity gain for the individual turns into only 1.7× for the organization.

Summary

Don’t ship your org chart to the user. The structure that makes a company easy to run is rarely the structure that makes a product feel like one product, and users experience the seams either way.
The Spotify model was never the problem — the copying was. What travelled was the diagram, not the culture underneath it, and the diagram is the part that doesn’t work on its own.
Autonomy stops scaling once dependencies grow geometrically. Every additional autonomous team adds more connections than it adds capacity, which is the point where local speed starts producing global slowness.
The Airbnb counter-example: deliberately moving from low to high coordination tax, on purpose, because a unified experience was worth paying for.
The Porsche stopped at a red light — team velocity is the wrong metric. A fast squad blocked on a dependency is still a stationary car, and the dashboard will still report it as fast.
Agents can’t work with the purposeful ambiguity operating models rely on. Humans quietly fill the gaps in a RACI; agents execute the gap literally, which exposes every unwritten assumption at once.
10× for the individual, 1.7× for the organization. The multiplier gets eaten by coordination on the way up — and the human ceiling on organizational complexity, the one that used to stop us from building anything more tangled, is gone.

What was discussed

What “operating system” actually means

Key insight

Hosanna and Nora define the operating model as the thing that decides how work actually moves, not the org chart that describes who reports to whom.
Their remit inside Spotify was the product operating system — the layer where hundreds of squads meet one product.

Don’t ship your org chart to the user

Key insight

The seams between teams show up in the product as inconsistency, duplication and dead ends.
Users don’t know or care which squad owns which surface — they experience one thing or they don’t.

Horizontals vs verticals

Key insight

The tension every scaling org faces — vertical teams that own outcomes against horizontal functions that own coherence.
Neither side wins permanently; the operating model is the ongoing negotiation between them.

The Airbnb story — low to high coordination tax, on purpose

Story

A deliberate move toward more coordination, accepted as the price of a unified experience.
Coordination tax isn’t automatically waste — it becomes waste when it’s unchosen and unmeasured.

The Spotify model — culture, not structure, was the problem

Key insight

What companies copied was the diagram; what made it work was the culture that never travelled with it.
The people who lived inside it describe something considerably messier than the version that got exported.

Autonomy at scale — when dependencies grow geometrically

Key insight

Adding autonomous teams adds connections faster than it adds throughput.
Past a certain size, more autonomy produces more coordination, not less.

The Porsche stopped at a red light

Highlight

Team velocity measures the engine, not the journey.
A squad can be fast and still deliver nothing if it’s blocked on someone else’s dependency.

Hard APIs between teams

Key insight

Treating team boundaries like interfaces — explicit contracts instead of implicit goodwill.
Getting systems of record in sync so the contract has something real behind it.

Why agents can’t work with purposeful ambiguity

Key insight

Operating models quietly depend on humans reading between the lines of a process nobody fully wrote down.
Agents execute what’s actually specified, which turns every unwritten assumption into a visible failure.

“Everyone arrives with 26 rocket ships”

Highlight

Hosanna’s image for what happens when every individual shows up with agent-scale capacity.
The constraint moves from how much can be produced to how much can be reconciled.

Productivity era vs outcome era for agents

Key insight

The first wave of agent adoption optimizes individual output.
The next one has to answer whether any of that output changed an outcome.

10× for the individual, 1.7× for the organization

Key insight

The gap between the two numbers is the coordination tax, paid on the way from personal output to company results.
The human ceiling on organizational complexity is gone — we can now build tangles no one could previously hold in their head.

Closing advice: “you can do anything, but you cannot do everything”

Highlight