A chrome wire grid floating over clear turquoise sea

AI in Strategy #2 · Helsinki

Do we need humans to do strategy?

Strategy is the work we've assumed machines can't touch: too ambiguous, too consequential, too human. That assumption is now a live question, and we turned the question into a map you can argue with.

AI in Strategy: an event series on what machine intelligence does to strategy work#2 hosted by Ville Tikka & Johannes Koponen · 19 August 2026

The provocation

Consider Phlebas, the first novel in Iain M. Banks's Culture series, is one of AI researcher Demis Hassabis's favourite books. In Banks's future, humans live extraordinarily rich lives, but when something truly difficult needs to be figured out, superintelligent machines called Minds do the strategic thinking.

What if Banks got that part right?

We tend to think of strategy as fundamentally human work: understanding the world, making sense of uncertainty, imagining possible futures, choosing between alternatives, and deciding what to do. But these are precisely the capabilities AI is getting better at.

An AI can already scan more information than a strategy team, generate more alternatives, simulate more scenarios, challenge more assumptions and update its view of the world more frequently than any annual strategy process. And this may only be the beginning.

First, the two questions

Every "can AI do strategy" argument splits into two arguments

One is about thinking, the other about trust. Strategy Science's special issue on the question keeps them apart, and so does this map: reasoning depth runs up the side, granted authority runs along the bottom.

Vertical axis

How deep is the reasoning?

  1. L1Predict. What happens if things continue? Reading the world as it is.
  2. L2Intervene. What happens if we act? Estimating the effect of a move before making it.
  3. L3Analogize. What if we did what they did? Borrowing logic from other worlds, no local data required.
  4. L4Invent. What future could we make true? Composing a model of something no data describes yet. Walmart before Walmart.

Horizontal axis

How much authority does it get?

  1. T1Inform. The machine watches and reports. You decide.
  2. T2Recommend. It proposes moves. You keep the pen.
  3. T3Operate. It acts on its own inside limits you set, and gets scored constantly.
  4. T4Execute. It commits the organization. The chair, not the analyst.

The map

Sixteen cells, three kinds of water

Clear water means AI already works there today. Deep water means the ability may exist but the trust infrastructure doesn't. Open sea means we genuinely don't know if machines can do it at all. Tap a cell.

AcrossMachine authority: how much the AI may decide on its own, from watching (T1) to committing the firm (T4).

UpReasoning depth: how hard the thinking is, from reading trends (L1) to inventing futures (L4).

Tap any cell
Clear water: live today Deep water: trust not built yet Open sea: capability contested

Reading the map

The frontier is L-shaped, not diagonal

Two tides, one empty corner

Autonomy and intelligence advance from different directions. Along the bottom row, machines win authority because their decisions get scored thousands of times: pricing, allocation, experiments. Verification makes delegation cheap, so the water clears from the left and stays clear deep into T3.

Up the left side, the deeper reasoning gets AI first as a helper: retrieval, sparring, copilots. Plenty of intelligence, almost no authority, because a new theory of the future can only be tested by betting the company on it.

So the map fills along two edges at once and the top right corner stays empty. That corner has a name in science fiction: in Iain M. Banks's Culture novels, the Minds run civilization end to end. Tap the corner cell to see the three separate locks on that door.