How AI is rewriting the decisions that leaders need to make

More than 50 years ago, Pablo Picasso remarked that “computers are useless. They can only give you answers.” His famous words reflected a world in which humans framed problems and computers merely solved them. AI is blurring that boundary.

AI’s influence now lies not only in the answers it generates, but in the frames from which those answers emerge. Frames direct organizational attention, shaping which possibilities become visible and which remain unseen. In this context, leaders are increasingly distinguished as much by the frames they compose as by the decisions they make with AI. Much of the greatest strategic value comes from recognizing possibilities that no one else has imagined.

The rapid deployment of AI is driven, across most organizations, by a simple value proposition: efficiency. They have embraced this logic at remarkable speed. In 2025, 88% of organizations deployed AI in at least one business function, up from 55% in 2023. Generative AI deployment surged from 33% to 71% in a single year. The greatest efficiency gains are rarely realised at the moment a decision is made. They arise further upstream, in the cognitive work of gathering information, developing insights and identifying opportunities. As AI increasingly supports these activities, it moves upstream into the processes that precede decision-making.

The strategic risk lies not in who makes the final decision, but in who shapes the thinking that precedes it. Organizations may continue making their own decisions while increasingly relying on AI to define the problem, narrow the range of plausible possibilities and influence which questions get asked in the first place.

Consider a business entering a new market. The strategy team uses AI to analyse trends, competitors, internal capabilities and customer demand. Within minutes, AI identifies promising opportunities, synthesises relevant evidence and highlights a small number of strategic directions. By the time the leadership team sits down, the discussion is no longer centred on defining the opportunity, but on evaluating the possibilities. This upstream role of AI gives rise to two distinct – and mutually reinforcing – dynamics: frame compression and frame convergence. Frame compression reduces exploration within organizations, while frame convergence reduces diversity across them.

Frame compression operates within organizations. Businesses once developed frames through iterative discussion, analysis and debate. AI compresses this process into minutes, changing more than the pace of decision-making. It alters the very economics of deliberation. By making coherent frames almost costless to generate while leaving their critical evaluation just as demanding, AI changes the incentive structure of deliberation. As a result, organizations have less incentive to question the assumptions underpinning those frames or to explore alternatives. Once a plausible frame emerges, competing interpretations can come to appear as unnecessary delay rather than valuable inquiry.

Frame convergence, meanwhile, emerges across the many organizations using the same foundation models and methods for processing data. Attention often focuses on whether those using AI possess different data. Yet strategic diversity depends not only on data, but on the interpretive structures used to organize it. Different datasets processed through similar framing logics can still produce remarkably similar assumptions. The risk is that organizations develop similar blind spots. Errors, then, become correlated rather than independent and entire industries can become more vulnerable to shocks.

The familiar governance solution to “keep a human in the loop” is less reassuring than it seems. The critical question is not whether humans review AI's outputs, but whether they still influence how the problem is framed before AI begins generating answers. They may remain in the approval loop while quietly disappearing from the framing loop. Indeed, a human can rigorously stress-test three possibilities and never ask whether those were the right three to consider in the first place. What appears to be oversight risks amounting to little more than choosing from a predefined menu.

Better prompting does not fundamentally resolve this problem. Prompts refine a search within a frame, but they rarely question whether the frame itself is appropriate. Although AI can generate many alternatives within a chosen frame, it is less reliable at exposing the assumptions that define it. Even if future models become capable of proposing genuinely different frames, organizations may still favour the efficiency of the first plausible framing over the effort required to explore alternative ones.

Neither human oversight nor better prompting addresses the underlying issue: who determines the frame before AI does. And that responsibility remains a leadership responsibility. Leadership's distinctive contribution increasingly lies in composing the frame. It is not simply about selecting the picture, but also about determining what enters it, what remains outside it, and what meaning the organization derives from it. In practice, this implies creating the conditions under which alternative frames can emerge before AI begins narrowing the field of possibilities.

As such, leadership teams need to create the conditions for robust strategic framing. These conditions start with questioning the problem before seeking answers, creating time to explore alternative frames, rewarding genuine challenge and resisting premature closure by treating the first plausible frame as a hypothesis rather than a conclusion.

Ultimately, this is not just a question of performance but of organizational sovereignty. Those that lose control of their framing gradually lose the capacity to define their own strategic agenda – and it is only those at the top who have the power to prevent this from happening.