The AI Absorption Gap

So, Dario Amodei is now calling for a slowing of the pace of development at the AI frontier, with the others giving their blessing to the proposal.

Setting aside what’s in it for them - yes, it would give them an opportunity to consolidate their positions - and the always valid questions around safety, given that capability is advancing faster than our ability to understand, control and govern increasingly powerful AI systems, I’m interested in looking at the proposal from the point of view of: what’s in it for us?

 Reframing the safety point a little, I think there is another strong reason why a temporary plateau in AI development could be useful.

 Right now, the world has more AI capability than it currently knows how to absorb.

 The trend is to measure AI progress by what happens inside the models: better reasoning, more capable agents, higher benchmark scores and greater autonomy.

 But technological capability is only one part of progress.

 The other is our ability to turn that technology into useful organisational and societal capability.

 McKinsey’s 2026 AI Trust Maturity Survey found that only around 30% of organisations had reached higher levels of maturity in areas including AI strategy, governance and agentic AI controls. It also found that nearly 60% identified knowledge and training gaps as a leading barrier to responsible AI adoption.

 The models have been moving very quickly, with organisations effectively chasing them and trying to adopt new capabilities almost as fast as they appear.

 If frontier AI capability barely improved over the next one or two years, I don’t foresee innovation stopping at all.

 Businesses would still be redesigning processes.

 Governments would still be developing regulation.

 Security teams would still be learning how to control agents.

 Schools and universities would still be adapting to different approaches to cognitive assessment.

 And organisations would still be trying to turn impressive AI demonstrations into measurable productivity and economic value.

 Advances in innovation would simply move elsewhere in the ecosystem. Instead of happening primarily inside the models, more of it could happen around them.

 AI capability isn’t the problem right now.

 Our ability to absorb it is.

An interim plateau in capability would give businesses, governments and society something increasingly valuable: time to catch up.