France's competition regulator just put a number on something enterprise IT leaders have felt anecdotally for the better part of two years: the AI agent market is concentrating fast, and it's concentrating around a very small number of companies. On July 17, 2026, the Autorité de la concurrence published Opinion No. 26-A-05, the product of a market inquiry it opened on January 8, 2026 into the competitive functioning of the artificial intelligence agents sector. The headline finding is stark — OpenAI, Google, and Anthropic together hold more than 84% of the global AI agent market. For any organization currently choosing, or already committed to, an AI agent vendor, that number is worth sitting with, because AI agent market concentration isn't an abstract policy concern. It's a direct input into how much leverage you'll have in your next vendor negotiation.
What the Autorité de la concurrence actually found
The scale of the inquiry itself signals how seriously French regulators are treating this. The final opinion runs to more than 3,700 pages once its annexes are included, and notably, the regulator didn't rely solely on submissions and testimony from the companies under review. As part of the inquiry, the Autorité de la concurrence built its own AI agents to study the sector empirically — an unusual and fairly aggressive move for a competition authority, and one that suggests the conclusions weren't reached lightly or based purely on vendor self-reporting.
The opinion's central concern is that the shift from conversational chatbots to autonomous agents — systems that don't just answer questions but take actions, call tools, and complete multi-step tasks on a user's or organization's behalf — could concentrate the digital economy around a small number of vertically integrated firms unless regulators act quickly. The report is explicit about the risks it sees materializing: "platformisation," where a handful of agent platforms become the default gateway through which businesses and consumers interact with software and services; disintermediation, where agents route around existing intermediaries and replace them with the agent provider itself; and algorithmic collusion, where pricing or behavior converges across supposedly independent AI systems in ways that harm competition without any explicit agreement between firms.
Beyond OpenAI, Google, and Anthropic, the report also maps the rest of the field — vertically integrated firms like Amazon, Microsoft, Meta, and Nvidia that are building agent capabilities into broader platform and infrastructure businesses, alongside specialist players such as Mistral AI, Perplexity AI, and xAI that are trying to compete without the same scale of underlying infrastructure or distribution. The regulator's read is that this second tier exists, but it isn't close to challenging the concentration at the top.
Why this matters even though it's non-binding
It's important to be precise here: Opinion No. 26-A-05 is exactly that — an opinion, not a binding regulation. The Autorité de la concurrence has no power in this document to force interoperability, break up a vendor, or mandate data portability. What it calls for is full enforcement of existing competition rules, interoperability requirements, and open standards — recommendations aimed at French and EU policymakers, not immediately enforceable obligations on OpenAI, Google, or Anthropic. If you're an IT leader wondering whether this opinion changes your contract terms next quarter, the honest answer is no, not directly, and not soon.
What it does do is add another data point to a broader pattern of EU regulatory scrutiny of AI and platform gatekeepers that's been building through 2026, including enforcement activity under the Digital Markets Act against companies like Apple and Google. Opinions like this one tend to function as groundwork — they shape the political and evidentiary basis for future binding rules, they give the European Commission and national regulators language and data to draw on, and they signal to the market which behaviors are likely to draw scrutiny even before any rule requires a change. Enterprise buyers should treat this as an early warning system, not an immediate compliance requirement. Track it, but don't restructure a live contract around it yet.
The practical lock-in problem this opinion is describing
Set the regulatory mechanics aside for a moment, because the underlying market dynamic the Autorité de la concurrence is describing is one that procurement and IT strategy teams are already living with. When three vendors control the overwhelming majority of a category, the practical difficulty of running a genuine multi-vendor agent strategy goes up sharply. AI agents aren't like swapping one API-compatible chatbot for another — an agent's value comes substantially from its tool and connector ecosystem: the integrations it has with your CRM, your ticketing system, your internal knowledge base, your identity provider, your document stores. Those integrations are frequently proprietary, built specifically around one vendor's agent framework, and non-trivial to reproduce elsewhere.
That's the mechanism by which AI vendor lock-in actually happens in practice. It's rarely a single dramatic moment where you're told you can't leave. It's a slow accumulation of custom connectors, fine-tuned prompts, workflow automations, and staff expertise that are all built around one provider's specific agent architecture. Eighteen months in, migrating away isn't a technical inconvenience — it's a project with its own budget line and its own risk of business disruption. Enterprise IT teams that have lived through cloud vendor lock-in in the 2010s will recognize the shape of this immediately: proprietary APIs, egress friction, and deeply integrated managed services that made switching clouds technically possible but organizationally expensive enough that most companies simply didn't. AI agents are on track to reproduce that pattern, potentially faster, because agent ecosystems are being built out and adopted at a pace that outstrips how carefully most organizations are documenting their own dependencies.
Why interoperability and open standards are the concrete fix
This is why the opinion's call for interoperability requirements and open standards isn't just regulatory boilerplate — it maps to specific, actionable concerns for enterprise buyers. Data portability means being able to export the context, memory, and configuration an agent has built up about your organization and reload it into a different system without starting from zero. Interoperable APIs mean an agent built on one vendor's platform can call tools and services defined in a common, non-proprietary format, rather than requiring bespoke integration work per vendor. Open standards for how agents describe their capabilities, authenticate to external systems, and exchange task state would let organizations mix agents from different providers for different jobs, rather than committing an entire workflow to a single vendor's stack.
None of that exists in a mature, widely adopted form yet across the agent ecosystem. That's precisely the gap the French regulator is flagging — and precisely the gap that determines how much real choice enterprise buyers will have three to five years from now.
Default placement and disintermediation, in practical terms
The report's warnings about "platformisation" and disintermediation aren't abstract either — they describe something enterprise IT teams are already navigating in miniature every time they decide which AI agent gets embedded as the default assistant across an organization's tools. When an agent from one of the three dominant vendors is set as the default entry point into email, documents, scheduling, and internal search, it doesn't just handle those tasks — it becomes the layer through which employees discover and invoke every other tool in the stack. Third-party software that isn't a first-class citizen in that agent's connector ecosystem effectively becomes harder to reach, regardless of its own merits. That's disintermediation in miniature: the agent provider inserts itself between the user and every other vendor relationship the organization has, and over time, usage patterns and internal tooling decisions start to bend around whichever agent has default placement, not necessarily whichever tool is best for a given job.
This is also where algorithmic collusion risk becomes more than a theoretical antitrust concern for buyers. If pricing, feature gating, or recommendation behavior across the dominant agent platforms starts to converge — not through explicit coordination, but through similar training data, similar optimization incentives, or similar commercial partnerships — enterprise buyers lose one of the main benefits competition is supposed to provide: genuine price and feature differentiation to choose between. Procurement teams evaluating multiple agent vendors side by side should watch for this specifically. If three competing bids look suspiciously similar in structure, pricing tiers, and default restrictions, that's worth flagging internally as a market-structure risk, not just accepting as the going rate.
What IT leaders should actually do now
Given that this is a non-binding opinion and not an enforceable rule, the right response isn't to pause AI agent adoption or wait for regulatory clarity that may be years away. It's to build the same discipline into agent procurement that more mature organizations already apply to cloud contracts. A few concrete moves matter here.
First, insist on interoperability and data portability language in vendor contracts now, while you still have negotiating leverage — before deep integration makes you a captive renewal rather than a prospective customer. Ask vendors directly what happens to your configuration, connectors, and accumulated agent context if you leave, and get the answer in writing. Second, avoid committing your most business-critical workflows to deep, single-vendor integration without an explicit exit plan — not a hypothetical one, but a documented estimate of what migration would actually cost and how long it would take. Third, track how many of your organization's agent-dependent workflows rely on genuinely proprietary connectors versus more portable, standards-based integrations, and treat that ratio as a risk metric your architecture and procurement teams review on a recurring basis, not a one-time assessment.
Finally, keep watching this regulatory thread, even though it moves slowly. Opinions like Opinion No. 26-A-05 tend to precede binding rulemaking by a year or more, and the specific remedies regulators eventually land on — mandated interoperability, data portability requirements, restrictions on default placement — are worth anticipating in your architecture decisions now, so you're not retrofitting compliance into a system that was built assuming none of it was coming.
The 84% figure from the Autorité de la concurrence isn't a prediction of where the AI agent market is headed — it's a measurement of where it already is. Enterprise buyers who treat that as a planning input today, rather than a policy footnote to revisit later, will have considerably more flexibility when the market — or the regulators — eventually force the interoperability question that vendors have so far been able to avoid answering.