Team & org6 min read

Two Flywheels: A Management Model Built for AI Startups

A structure I stumbled into on my first internship — and have been thinking about ever since.

By Alex · AirJelly team

My first internship was at a Wyoming AI startup. My title was AI Research, but in practice I was doing AI PM work.

One day, I asked to move from the big office to the small one. The reason was simple: the big-office crowd didn't show up until the afternoon, I was already there every morning, and I felt awkward leaving early. The small-office people came in earlier, so I figured that if I moved there, I might be able to leave earlier too.

That one unremarkable decision led me to something I've been thinking about ever since.

Two offices, two timelines

After I moved, I realized the two offices were not just physically separate. They were living on entirely different timelines — not in terms of working hours, but in terms of which version of the company each group was inhabiting.

Say the current product was at v1.0. The big office was running v1.0 and building v2.0. The small office was drafting v4.0 and conceptualizing v5.0.

A bit of context: the big office was made up of experienced full-timers on established teams, focused on build and support. The small office was all AI PM interns, and our job was to explore.

A deliberate misalignment

It's counterintuitive. Most teams — including the near-unicorn I'm at now — optimize for alignment. This structure did the opposite: it deliberately preserved a kind of misalignment.

If explore and build sit in the same room, explore gets swallowed by day-to-day handoffs and coordination. Urgent work almost always gets handled before important work.

The urgent-versus-important matrix has trained many of us to treat urgency as the more legitimate form of work.

The roles nobody assigned

Then I noticed something else — or, more accurately, something else grew on its own. There were two interns in the small office before I joined, three after. Officially, our job was to give the founder inspiration. In practice, the founder was the biggest PM in the company.

Because our time was flexible and the three of us were wired differently, we slowly fell into a rhythm I still miss. None of these roles were assigned in advance. They emerged from the work itself, and they worked unusually well.

01Role

The Scout

Tracks the latest frontier AI developments and translates them into outputs the rest of the team can actually understand.

↓

02Role

The Systematizer

Engineers those raw outputs into SOPs anyone in the company can pick up and run.

↓

03Role

The Bridge

Connects the frontier back to the big office, and carries real frontline community feedback back to the small one.

Frontier → systems → frontline

The takeaway

One office holds the present. One runs toward the future.

Big officeHolds down the present

Small officeRuns toward the future

So if you're building an AI startup, or your company is going through an AI transformation, this structure is worth trying. It doesn't require much extra headcount, resources, or organizational design. Inside the small office, three roles tend to emerge naturally: someone who scouts the frontier, someone who turns it into systems people can use, and someone who connects the frontier back to operational reality.