Help your leaders make better decisions about AI.

For three years the through line was simple: AI kept getting better and cheaper. That still holds at the floor, where capable models cost cents per million tokens, though the frontier itself can now get pricier again. The deeper change is different in kind. The length of work these systems finish reliably is doubling every few months, and that pace is still speeding up. We are living in the exponential, while most organizations still plan as if the curve were a straight line. AI has stopped being something you prompt and become something that works, and lately, something that improves itself, helping train and design its own successors.

The craft moved with it, from writing prompts, to giving models context, to setting self-running loops that carry a task to done. Agents now handle multi-step work unattended, and they are starting to replace the software your teams log into and the coordination layer that holds an organization together. As one person learns to direct a swarm of them, the scarce skill stops being execution and becomes judgment: knowing what to ask for, and seeing when something is off.

Two gaps decide who wins. The first is between what these models can already do and what most companies actually ask of them, still optimizing old tasks instead of asking which tasks should exist at all. The second is between the firms that consume AI and the firms that control it, between those who rent intelligence and those who own strategic access to it. A growing share of the world's AI capability now sits inside a handful of companies and jurisdictions, and a model can appear and vanish overnight, so the real question becomes how dependent you are willing to be on capability you do not own.

That is why I keep coming back to AI sovereignty on stage. Open models keep improving, compute keeps getting cheaper, and the ability to run powerful models on your own hardware has gone from a distant idea to a real strategic option. The organizations building their own AI capability, literacy, and sovereignty now are not chasing efficiency, they are buying resilience. The gap between them and the firms that only consume AI is widening fast, and it is the kind of gap laggards may never close.