Capability Overhang: The AI We Haven't Used Yet

Ethan Mollick’s latest Substack article, The Overhang (link in comments), gave me a term I now find myself using: capability overhang.

It’s the gap between what today’s AI models can already do and what almost anyone is actually doing with them.

Most of the conversation and worries around AI right now are about what the next generation of models might be able to do. But a question that’s hiding in plain sight is: do we even know what the models we already have are capable of?

He describes experiments where current models take on far more ambitious tasks than expected, using tools and approaches that surprised even experienced users. His larger argument is that if AI development stopped tomorrow, the unused capability in existing models could still drive substantial change.

And that makes capability overhang an opportunity. The people and organizations that learn to explore this gap, combining AI with their own expertise, context, taste and agency, may get dramatically more out of the same models everyone else has access to.

The next big AI capability may not need to be invented. It may already be here, waiting for someone to use it.