When running an analysis cost a person two days, you ran the analysis you were fairly sure you needed. When it costs an agent twelve minutes, you can afford the one you're merely curious about.
The practical cost of testing a hypothesis — not the theoretical cost, but the calendar time and headcount it consumed — worked as a filter. Teams investigated what they could justify investigating. Whether the pricing model that closed deals in healthcare would work in logistics. Whether the architecture could absorb the integration a key customer kept asking for. Questions like these surfaced in planning meetings and then quietly died, because nobody could defend spending a week on them alongside the questions the organization had already committed to.
Cheap agent execution loosens that filter. A scenario-planning comparison across three firms found that managers working with an AI facilitator identified nearly twice as many external driving forces and strategic options as a control group at a whiteboard, inside the same four-hour workshop. Both groups converged on three final scenarios. The AI-assisted group had simply walked through more of the space before converging. The inputs to the decision expanded while the decision's shape stayed the same.
A product manager at a software company described turning a research paper on failure classification into a working prototype in under two hours, against an estimate of one to two days by hand. The prototype wasn't the point. At two hours, building it is a reasonable way to find out whether the idea is any good. At two days, you need a case before you spend the time. What moved was the threshold for justifiable curiosity.
Sable Whitford argued recently, writing about articulation work, that the hours automation gives back tend to get absorbed by specifying and maintaining agent behavior rather than freed for strategy — a useful corrective to the assumption that automation lifts everyone's work by default. Fair enough. But in teams that have already paid down that maintenance cost, something else shows up: they start asking questions they previously couldn't budget for.
More exploration does not automatically produce better decisions. A randomized experiment with 305 MBA students found that AI assistance increased the number of strategic alternatives generated. But when participants brought the AI in during problem formulation, rather than waiting until later ideation, their alternatives came out less strategically focused. The tool helped them see more possibilities while quietly shaping how they framed the problem in the first place. They explored more widely but framed more narrowly.
That result deserves more attention than it has gotten, because it identifies where the skill actually lives. In an environment where options are cheap, the difficult thing is holding on to which question you are trying to answer while the space of possible answers expands around you. Teams that explore well will be the ones that can keep their problem definition steady under that pressure — or revise it on purpose, knowing they're doing it, rather than drifting into a reframing the tool suggested.
Affordability used to be the thing that stopped you: we can't run that analysis. Now you can run thirty of them, and the difficulty moves downstream — thirty results, and someone has to work out what they collectively say. Which, if you've read the companion piece, starts to sound like composition.