The vision · The decade ahead

The next decade of humans and AI

I will make a falsifiable bet. The next ten years of AI will be decided less by how capable the models become and more by whether we build the unglamorous infrastructure that keeps a human genuinely in command of them.

The next decade of humans and AI will turn on command, not capability. Capability will keep climbing, and it will matter less than the headlines suggest, because the decisive question becomes whether a person can still understand, override, and answer for what these systems do. That is not a model problem. It is an infrastructure problem, and the human judgment infrastructure to solve it barely exists yet. Building it is the bet I am making with my time.

Less about capability, more about command

The early era of AI put the human at training time, labeling data. The era arriving puts the human at run time and at the level of the rules, deciding in the moment and deciding the constraints. The center of gravity of the work moves from making models capable to keeping humans in command of capable models. The teams that see this early will build the layer everyone else ends up needing.

The infrastructure that is missing

We have benchmarks for capability and almost nothing for command. No shared measures of whether oversight is real, no standard way to route a hard decision to the right human in time, no agreed account of who is accountable. That absence is the opportunity. The boring layer, measurement, escalation, accountability, is the one that decides whether the next decade is something we can live with.