Radical federalism: The way to tackle the AI velocity problem

The AI velocity problem

Organizations plan in three- to five-year cycles, but AI evolves in six- to 12-month cycles, creating a fundamental mismatch. Traditional planning approaches – centralized control on the one hand and chaotic adoption on the other – fail to address this reality. 

The third way: radical federalism

The third way, which I refer to as radical federalism, pushes adaptive capacity to the unit level while maintaining minimal coordination centrally. Each department designs its own human-AI configuration based on its specific variance, velocity, and local knowledge. This temporal fracturing, combined with high variance and local knowledge requirements, makes federalism not just useful but necessary. The operating model follows four concrete principles: 

1. Budget autonomy, not headcount

Give departments budgets and let them decide their optimal mix of people, AI subscriptions, and hybrid approaches. 

2. Outcome accountability, not process compliance

The center defines what to deliver, not how to deliver it. 

3. Interface standardization, not tool standardization

Different departments can use different AI platforms as long as data exchanges cleanly. 

4. Local iteration speed

Each department reorganizes at the pace its domain requires.