Lore

AI Needs Heavy Constraint to Be Useful for Strategy

Contrary to his own expectation, Adam Mosseri (Instagram) found AI is not naturally good at originating strategy. Loosely-prompted AI strategy output is predictable and generic; it only becomes useful when aggressively constrained with real inputs — the actual state of the technology, the personnel available, the competitive/regulatory/compliance landscape, and brand identity. His explanation: good strategy has to be 'controversial' — genuinely disputable, cutting against some plausible alternative — and a model trained toward broadly agreeable output defaults to the safe, average answer unless forced away from it by hard constraints. He draws the implication that as AI automates execution, human judgment concentrates increasingly in vision and strategy, and steering AI toward a non-generic answer becomes its own management craft, distinct from just prompting well. Complements Taste as the Human Differentiator and product-leadership-as-curation.

Instagram/Mosseri Example

Adam Mosseri (Instagram) says he assumed AI would be especially good at strategy work, since a model can be given access to market data, competitor moves, and internal metrics. In practice it wasn't: left unconstrained, the model produces the same predictable answer competitors would already expect. Useful strategic output only came from deliberately narrowing the model with concrete, specific inputs rather than letting it reason broadly over all available data.