France’s competition regulator has done something unusual: it built its own AI agents, gave them 550 shopping tasks, and recorded exactly which companies and websites they visited. The result is a 3,700-page opinion that should be required reading for any business leader deploying or depending on AI agent technology.

On July 17, 2026, the Autorité de la concurrence published Opinion 26-A-05 on the competitive functioning of the AI agents sector. The headline finding is stark: OpenAI, Google, and Anthropic together control more than 84% of the global AI agent market as of May 2026. Three companies. Eighty-four percent. A sector that barely existed in its current form two years ago.

What the Regulator Actually Found

The Autorité’s opinion is the third time France’s competition watchdog has weighed in on the AI value chain. Its previous opinions looked upstream at model training. This one focuses on the deployment layer, where AI agents actually execute tasks, make decisions, and increasingly handle commerce on behalf of users.

The regulator identified three categories of competitive risk that it says could harden the market around a small number of vertically integrated firms unless regulators and businesses act quickly.

Platformisation is the risk that a handful of AI agent platforms become the default gateway through which users access the entire digital economy. Much like search engines once became the entry point to the web, AI agents are on track to become the interface through which consumers and businesses discover services, compare prices, book appointments, and make purchases. When one agent vendor controls that gateway, the implications for competition are significant.

Disintermediation concerns what happens to the businesses currently sitting between consumers and outcomes. If an AI agent can complete a purchase without a user ever visiting a comparison site, a marketplace, or even a brand’s own homepage, the entire layer of discovery and consideration collapses. Businesses that depend on being found get skipped entirely.

Algorithmic collusion is perhaps the least discussed of the three risks, but the one with the most far-reaching implications. When competing businesses all use AI agents trained on similar data or built by the same underlying model providers, those agents may converge on similar pricing, negotiating, and selection strategies without any explicit coordination. The result can look like collusion without any of the human actors intending it.

The Regulator’s Recommendations

The Autorité is not calling for breaking up the big AI providers. Instead, the opinion calls for full enforcement of existing competition rules, mandatory interoperability between AI agent platforms, and open standards that prevent lock-in.

Specifically, the opinion recommends that AI agents should be required to work across platforms rather than being siloed inside a single provider’s ecosystem. A business that builds workflows around one AI agent vendor should not be penalized with switching costs and broken integrations if they choose to change providers.

What This Means for Business

The practical implications run in two directions.

First, if you are building business operations around AI agents right now, vendor lock-in is a genuine risk. The French opinion mirrors concerns raised by enterprise architects and procurement teams globally: the costs of switching agent platforms may become prohibitive as integrations deepen. Businesses that build with interoperability in mind today will have more flexibility tomorrow.

Second, if your business depends on being discovered, compared, or recommended, agentic commerce changes the rules. AI agents that handle purchasing decisions on behalf of users create entirely new categories of winner-takes-most dynamics. Being present in the training data and tool-use context of the dominant agent platforms will matter as much as SEO rankings do today, potentially more.

The French findings also signal that regulatory scrutiny of AI agent providers is only going to intensify. Businesses in regulated sectors should be planning for compliance requirements around AI agents well before those requirements arrive.

The window to make strategic choices about AI agent adoption, vendor selection, and workflow design is now. The companies that move thoughtfully in the next twelve months will be far better positioned than those who wait for the regulatory landscape to fully clarify.

---

Enterprise DNA helps businesses evaluate, adopt, and build AI agent strategies that do not create dangerous dependencies. Talk to our advisory team about building AI operations that work regardless of which model provider wins the platform wars.