Adoption doubled. Profit did not.
Large companies expanded their agent rollouts by nearly half. The share reporting any impact on profit stayed exactly where it was.
The gap between companies running agents and companies earning from them is not a technology problem. It is an ownership problem.
1. Someone finally measured the implementation gap
In a survey of 1,719 leaders, the share of large companies scaling AI agents rose from 27% to 40%. The share reporting any operating-profit impact stayed at 37%, and only 6% qualified as AI high performers.
Source: McKinsey State of AI survey ↗Take one agent already in production and put it on a P&L line—with a named owner and a number. Without an owner for the number, it is still a pilot.
2. The buying decision is becoming a build decision
Nearly a third of respondents said they chose to build a capability with agentic development tools instead of buying a software feature. Software procurement is no longer only a price negotiation; it is a comparison between buying and building quickly for the exact need.
Source: The finding in McKinsey’s survey ↗Before the next major software renewal, ask the team to price the cost of building the three capabilities the company actually uses.
3. When a core capability is missing, time matters more than building
Vanguard’s agreement to acquire Altruist illustrates what large companies buy when they do not want to wait: infrastructure, expertise, and years of execution already completed. The CEO question is not only what can be built, but what the company cannot afford to build slowly.
Source: Vanguard announcement ↗Name one strategic capability the company will not build itself, then begin pricing an acquisition or partnership this quarter.
4. Agents are moving from screens into the physical world
Standards for agents operating physical equipment open a new front in laboratories, factories, and warehouses. The advantage will not begin with the model. It will begin with the company that already documented the workflows around its equipment.
Source: Anthropic research preview ↗Choose one workflow around one instrument and start logging actions, exceptions, and results. That data will become the raw material for implementation.
5. The control layer is part of the product
As agents receive more authority, trust, safety, and control move from overhead to operating system. A company that funds the agent but postpones governance creates operational debt from day one.
Source: Alice funding announcement ↗Fund the control layer from the same budget line as the agent and approve both in the same meeting.
The big picture
Companies creating value from AI are not simply buying more tools. They connect every implementation to a business owner, an operating metric, and a control layer. The technology is only the beginning of the work.
One useful thing this week
Choose one AI implementation already inside the company. Write down three things: who owns it, which number should change, and when that number gets reviewed. If one answer is missing, that is this week’s job.