McKinsey & Company has published a piece titled "The agentic advertising economy: From attention to action," laying out its view of where digital advertising goes next once AI agents start doing the buying.

The core claim sits in the title. Advertising today runs on attention: impressions, clicks, viewability, brand recall. McKinsey's argument is that this model erodes once consumers delegate tasks to AI agents that compare products, negotiate prices, and complete purchases on their behalf. The unit of value shifts from capturing a human's attention to prompting an agent's action.

That is the concept. What does it mean in practice for operators and marketers?

From selling to persuading software

If an agent shops for a flight, a phone plan, or a subscription, the ad is no longer a message aimed at a person. It is structured information an agent can parse — price, terms, availability, delivery windows, warranty conditions. The persuasion target becomes the algorithm that ranks options, not the buyer skimming a feed.

This reframes several parts of the current stack. Creative work, media buying, and attribution all assume a human at the other end of the funnel. In an agent-mediated purchase, the funnel compresses to a query and a decision made in milliseconds by software weighing stated criteria.

What the shift implies

McKinsey's framing suggests a few operational consequences worth watching:

• Product data becomes advertising. Clean, complete, machine-readable feeds may matter more than creative assets. If the agent cannot parse an offer, the offer effectively does not exist.
• Price and terms face direct comparison. Agents compare exhaustively and without brand loyalty. Margin pressure on undifferentiated offers likely intensifies.
• Distribution shifts to agent platforms. Whoever owns the agent — an OS vendor, a browser maker, an assistant app — controls the shelf. That concentrates power in a new gatekeeper layer, echoing earlier fights over search and app stores.
• Attribution breaks. Last-click and view-through models assume observable human behavior. Agent-initiated transactions need new measurement logic.

Skeptics will ask the obvious questions

McKinsey is not the first consultancy to coin an economy. "Agentic" is currently among the most overworked words in the vendor lexicon, and readers should weigh this piece accordingly: it is a framing document from an advisory firm whose business includes selling transformation work, not a product announcement or a dataset.

Real-world friction also remains. Consumers have historically been slow to hand full purchase authority to software. Agents need permission structures, error handling, and trust mechanisms that mostly do not exist yet at scale. Regulatory scrutiny of AI intermediaries — and of who pays whom for placement inside an agent's results — is a live question on several continents.

Still, the direction of travel is plausible. Search behavior already skews toward summaries and zero-click results. Commerce APIs and comparison tooling have trained the market for machine-readable offers. The agent layer is the logical next step, and advertisers who treat product data as a second-class asset will feel it first.

The takeaway

No ship date, no pricing, no named customers — this is a conceptual piece, not a launch. But the underlying signal matters to anyone who sells through digital channels: prepare offers that software can read, terms that survive side-by-side comparison, and measurement that does not depend on a human clicking an ad.

Attention funded the last two decades of the internet. McKinsey is betting the next one runs on action.