Most companies haven't decided how to use AI — mandate vs redesign

Two CEOs used AI to cut headcount dramatically. They made opposite bets to get there.

Tobi Lütke at Shopify issued a mandate: prove that AI cannot do the job before asking to hire a human. The org chart stayed roughly intact. AI became a forcing function before any headcount decision. Headcount is down around 30%, revenue is up more than 20% annually. He didn’t redesign anything. He inserted a constraint.

Jack Dorsey at Block did something structurally different. He threw out the org chart and rebuilt the operating model from scratch. Four layers replacing traditional hierarchy: capabilities, world models, intelligence, interfaces. Three roles replacing management: builders, outcome owners with temporary authority, and player-coaches. 4,000 jobs cut in February 2026. Not as a cost measure, as the first move in a new model.

The surface result looks similar: fewer people, more output per person, AI doing work that humans used to do. But the underlying bets are completely different.

Lütke’s model is evolutionary. The structure you had still exists. AI is additive pressure. You can reverse a mandate. It’s low disruption, high discipline.

Dorsey’s model is architectural. You can’t reverse a redesign while you’re running the company. It’s a different theory of what an organisation is, not a hierarchy for routing information, but a system for developing and deploying intelligence.

Neither is obviously right. But they’re not the same choice, and I think most leaders haven’t noticed that.

When I talk to founders and operators about their AI strategy, the conversation usually slides toward tools: which model, which workflow, which team is using it. The structural question, whether you’re inserting AI into an existing model or rebuilding around it, rarely comes up. Often because they haven’t decided. Or they’ve been making the mandate choice without calling it that, while talking about it like the redesign choice.

The two choices carry different consequences. A mandate preserves accountability structures and is reversible if it stops working. A redesign creates a vacuum you have to fill deliberately, and you can’t really undo it mid-flight. One asks whether AI can do the job. The other asks what a job is.

I’ve been building toward the redesign model from the start. My current setup is a multi-agent executive assistant, a COO layer, researchers, brand strategists, store operators, all agents orchestrated through a shared context and memory system. I run it alone. It works because the architecture doesn’t assume humans in those roles. It was designed around intelligence, not headcount.

The part I haven’t solved yet is making it multi-person. Right now it scales with me, not with a team. That’s the next problem, and I think it’s the interesting one: what does it look like when the redesign model meets a real organisation with real people in it?

Which model are you building toward?