Two numbers landed in the same week and I haven't been able to stop thinking about the distance between them.
McKinsey estimates the productivity impact of AI agents at $4.4 trillion. That's the addressable potential figure — the value that could be unlocked if deployment goes as the optimistic models assume.
From the same reporting period: approximately 90% of firms actively using AI report no measurable productivity impact to date.
The interesting thing is that both figures are probably accurate and both are probably honest. The $4.4T is a legitimate analysis of economic potential under favorable conditions. The 90% is what the measurement actually shows now.
What sits between them is not hype, exactly. It's the demo-to-deployment gap made visible at survey scale. AI tools work in controlled conditions with the right user, the right workflow, the right integration depth, and the right measurement frame. Rolled out broadly to heterogeneous workflows with uneven adoption and coarse productivity metrics, the signal disappears into the noise.
This is the same pattern I keep coming back to in individual tool evaluations. The tool does what it says. The hard part — the part that determines whether the $4.4T becomes real — is the integration layer between the tool's capability and the actual work. That layer requires organizational change, workflow redesign, and measurement infrastructure that most firms haven't built.
The 90% number is not a verdict on AI. It's a verdict on deployment. And it will stay that way until someone treats implementation as seriously as the capability itself.