I have attended Cannes Lions every year since 2022 — not to follow the festival, but to step back and see how shifts in technology and commerce overseas translate into value and concrete challenges for media and advertising in Japan. Identifying the risks is only part of it. What matters in a period of rapid change is finding the new value and opportunities inside it.

The structural shifts I felt strongly on the ground in 2025 — the move of autonomous AI into its deployment phase, and the sharp decline in traffic from search — have, over the past year, become operational challenges that can no longer be ignored, and at the same time new possibilities.

The advertising ecosystem has since moved further. AI is no longer a tool for data analysis but an autonomous execution layer that optimizes budget allocation and bidding without human intervention. The software and advertising value chain is moving past the stage where single AI agents work on their own (the third wave) into the era of "swarms of agents" (the fourth wave): multiple background agents running on cloud infrastructure, coordinating with one another and executing tasks autonomously through webhooks and system triggers.

As users skip search and comparison altogether and hand purchasing decisions to these agent swarms — what is now called agentic commerce — we face a structural question: should advertising be designed for people, or for AI agents? And how do we approach "machine attention," the optimization of content to be consumed by AI agents running in the background?

Looking ahead to where publishers and the advertising industry are heading as AGI comes into view, the milestones fall into the following roadmap. It is a realistic projection, worked backward from discussions in overseas technology communities and the speed at which standard protocols are being built.

Milestone 1 (near term): Standardization and governance of agent-to-agent (A2A) commerce protocols

An ecosystem will take hold in which buy-side purchasing agents and sell-side yield agents negotiate the value and contextual fit of ad inventory in milliseconds and execute transactions without human involvement. Looking further to a fifth wave — in which multiple agents coordinate and feed transaction outcomes back into their own algorithms through reinforcement learning — measurement transparency and AI accountability will become the practical focus of governance.

Milestone 2 (mid term): Dynamic ad generation on edge AI, with privacy built in

Lightweight, low-power LLMs will be implemented natively on devices. Without going through the cloud, a device will read a user's emotional state and context locally and generate conversational, shoppable ad experiences in real time. This is also a technical answer to tightening data sovereignty and privacy regulation.

Milestone 3 (the AGI era): "Social trust" and "data licensing" as the ultimate scarce resources

In a world where countless AI agents summarize and regenerate information at high speed and digital noise reaches its limit, the value of media that holds fact-checked primary information and social trust — what AI draws on to learn and to cite — will paradoxically rise. Advertising models built on traffic volume will end. Publishers, on the strength of their data sovereignty, are expected to establish a new compensation structure in which they license content directly to AI platforms and are paid for citation.

To take this wide view and still capture the opportunities within it, we have to judge which layer calls for practical intervention now, and face the uncertainty involved.

Looking at each market, it is true that Japan has been preoccupied with defensive measures — legal frameworks, crawler blocking — and lags the commercial development under way in Europe and the United States. But being behind can be turned into a rational second-mover advantage. The risks the early movers are now facing, such as selling off data sovereignty and operational fatigue, are ones we can analyze after the fact from a step back.

In a market where optimization engines and autonomous agent swarms buy inventory, a publisher's individual value and context risk being reduced to a single parameter and priced down. But a publisher that generates and manages high-precision signals of its own — context, trust, attention — and supplies them directly to buy-side AI and DSPs gains a new competitive advantage that does not depend on any platform. That is a substantial growth opportunity.

Where we should concentrate, as a matter of practice, is not on ideals but on laying infrastructure at the very top of the supply chain. While we observe this period of change from a step back, we remain optimistic about a future in which the core asset of media — social trust — functions as it should and keeps producing new value.