Closing the gen AI paradox: a leadership agenda
Most companies have deployed generative AI, and most report no impact on earnings. McKinsey calls this the gen AI paradox, and argues only the CEO can resolve it.
Seizing the agentic AI advantage opens with a contradiction most executives will recognise. Nearly eight in ten companies report using generative AI. Roughly the same share report no significant impact on the bottom line. McKinsey names this the gen AI paradox.
The gap is simple to state: the tools were adopted everywhere, and the work behind them was rarely redesigned.
Horizontal copilots, vertical use cases
The explanation is an imbalance between two kinds of deployment. Horizontal copilots and chatbots, which sit across the whole enterprise, scale quickly but deliver diffuse, hard-to-measure gains. The vertical, function-specific use cases that would actually move the P&L are far more transformative, and McKinsey reports that about 90 percent of them remain stuck in pilot.
McKinsey's prescription is that agentic AI is not an incremental tool but the foundation of a next-generation operating model, and that the pivot from experimentation to scaled impact is a move only the CEO can make. Some firms have begun, not by deploying more agents, but by rewiring the organisation around them.
What the CEO actually has to do
The framing can mislead if it sends leaders rushing toward fleets of agents. Horizontal copilots underdeliver because broad, optional tools change behaviour without changing the work. Vertical use cases stall for a different reason: they require redesigning a workflow, its data, and its controls, which is hard, and which is exactly why so few leave the pilot stage.
So the CEO's job is to force the harder choice: pick the two or three functions where AI could change the economics, fund the workflow redesign, and hold the effort to a P&L target. That is uncomfortable because it concentrates resources and creates accountability. It is also the only approach the data suggests works.
A leadership checklist
- Distinguish copilots from transformation. Copilots lift individual productivity; transformation changes a business outcome. Do not mistake activity for progress.
- Choose a small number of vertical use cases and resource them properly instead of spreading budget across enterprise-wide tools.
- Put a named owner and a financial target on each, and treat unblocking pilots as a board-level priority.
Related reading
Source: Seizing the agentic AI advantage, McKinsey.