As companies race to scale artificial intelligence, the harder question is whether growth signals progress, or mounting complexity.

When McKinsey’s CEO Bob Sternfels said at CES that the firm now operates with roughly 25,000 AI agents, he was describing more than a technology rollout. In this framing, AI is no longer just a set of tools, but part of the workforce itself: scaling faster than any hiring plan ever could. The logic is straightforward: firms that move quickly on AI should gain an edge over those that do not.

But speed and scale, on their own, have never guaranteed efficiency.

In large organizations, rising counts of systems, processes, or headcount often signal complexity rather than leverage. AI agents can accelerate output, but they can also multiply activity: more analyses, more drafts, more dashboards, without improving the quality or speed of decisions. Technology is then asked not to simplify the enterprise, but to absorb friction it has not yet resolved.

This raises an uncomfortable question. If a firm requires tens of thousands of agents to operate, is it achieving operating leverage, or compensating for structural complexity?

Sternfels has pointed to concrete productivity gains, noting that AI has already saved McKinsey an estimated 1.5 million human hours. That is a meaningful operational milestone. But at the firm’s scale (roughly 40,000 professionals) that savings likely represents on the order of ~2% of annual labor hours. Useful, certainly. Transformative, not yet.

This distinction extends beyond consulting. In organizations where labor and coordination drive costs, saving time does not automatically improve economics. Productivity becomes leverage only when it leads to real changes in how work is organized: lower costs, clearer decision rights, or faster, higher-quality decisions. Otherwise, efficiency gains simply increase activity without changing decisions, risk, or capital allocation.

From a board-level perspective, the relevant question is not how many agents a firm deploys, but where they operate in the value chain. Are they embedded in pricing decisions, risk assessment, capital allocation, and execution, areas where judgment matters and mistakes are costly? Or are they concentrated on the sidelines, improving convenience without changing outcomes?

The firms that benefit most from AI will not be those with the largest fleets of agents. They will be those that can point to a smaller number of systems that materially improve decision quality, compress turnaround times, and create real operating leverage.

In that light, 25,000 AI agents is not yet an achievement. It is a question. And the answer lies not in scale itself, but in whether that scale produces clearer decisions, stronger economics, and simpler organizations, rather than a more sophisticated way to manage complexity.