Competition Law and AI Agent Marketplace Competition

Competition Law and AI Agent Marketplace Competition

1. Introduction

AI agents are rapidly developing from ordinary chatbots into software systems capable of planning tasks, selecting digital services, communicating with other software, making recommendations, and carrying out transactions for users. In July 2026, the French Competition Authority specifically examined competition in the AI-agent sector, noting that agents increasingly perform reasoning, planning and orchestration functions and may coordinate with other agents.

This development creates the possibility of AI agent marketplaces: digital environments in which users or businesses can discover, install, connect, purchase, or authorize AI agents and related tools.

From a competition-law perspective, such marketplaces can resemble earlier digital platforms such as app stores, search engines and operating-system ecosystems. The central concern is that the operator of an agent marketplace may simultaneously:

• operate the marketplace;
• establish access and ranking rules;
• control technical interfaces and APIs;
• provide the underlying AI model or assistant;
• distribute its own competing agents;
• control user data;
• process transactions; and
• decide which third-party agents can reach users.

This combination can create significant gatekeeper power.

There is still limited decided case law dealing specifically with AI-agent marketplaces. Therefore, existing cases concerning operating systems, search engines, app stores, digital marketplaces, interoperability and platform self-preferencing provide the most useful legal analogies.

2. Meaning of an AI Agent Marketplace

An AI agent marketplace can be understood as a platform connecting several groups:

Agent developers → Marketplace → AI assistant/user → Third-party services

For example, a user might ask a general AI assistant:

“Find me the cheapest suitable hotel and book it.”

The assistant could search a marketplace of specialized travel agents, select one agent, allow that agent to communicate with booking services, and return the result.

The marketplace therefore becomes more than a software catalogue. It may determine which agents receive access to users and which services those agents can reach.

This intermediary position is important because control over discovery can become almost as important as control over the underlying technology.

3. Relevant Markets

Competition authorities normally begin by identifying the relevant product and geographic markets.

Possible product markets in an AI-agent ecosystem could include:

• AI agent distribution marketplaces;
• general-purpose AI assistants;
• specialized AI agents;
• foundation models;
• agent-development platforms;
• cloud AI infrastructure;
• AI APIs;
• agent identity and authentication services;
• agent payment systems;
• agent-to-agent communication infrastructure.

The correct definition would depend on evidence concerning substitutability, pricing, functionality, switching behaviour and entry barriers.

A major issue will be whether an AI-agent marketplace constitutes a separate market or merely one distribution channel within a broader software market.

4. Network Effects

AI-agent marketplaces can develop strong network effects.

More users attract more developers because developers want access to those users.

More developers create more agents and functionality.

More agents make the marketplace attractive to additional users.

This creates a feedback loop:

More users → More developers → More agents → Better marketplace → More users

Network effects are not inherently unlawful. They become relevant where they contribute to durable market power or make entry by competing marketplaces unusually difficult.

5. Data Advantages

Marketplace operators could obtain substantial information concerning:

• which agents users select;
• successful and unsuccessful tasks;
• transaction patterns;
• user preferences;
• agent performance;
• conversion rates;
• pricing;
• developer activity.

A vertically integrated marketplace could potentially use such information to improve its own competing agents.

Competition authorities may therefore examine whether proprietary marketplace data creates an entry barrier or whether commercially sensitive third-party data is being used to disadvantage marketplace participants.

6. Self-Preferencing

One of the most important issues is self-preferencing.

Suppose Marketplace M distributes agents developed by independent companies but also owns Agent X.

When someone searches:

“Find an accounting agent.”

Marketplace M could technically place Agent X above competing agents.

Potential preferential mechanisms include:

• higher rankings;
• preferred recommendation slots;
• lower commissions;
• exclusive access to APIs;
• faster execution;
• privileged user data;
• default installation;
• better interoperability;
• preferential security approvals.

Competition law would examine the surrounding market power and whether such treatment constitutes exclusionary conduct under the applicable legal system.

7. Default Agent Selection

Defaults can have substantial competitive importance.

Imagine that a smartphone or operating system automatically directs every user request through one particular AI agent.

Users technically might be able to replace it, but switching could require multiple configuration steps.

If most consumers retain defaults, controlling the default can become a powerful distribution advantage.

This issue closely resembles earlier competition disputes concerning default browsers and search engines.

8. Exclusive Agreements

An AI marketplace might enter agreements requiring popular developers to distribute agents exclusively through its platform.

For example:

Developer A → Marketplace X only

rather than:

Developer A → Marketplaces X, Y and Z

Exclusive distribution can sometimes have legitimate commercial explanations, such as financing development or protecting investments.

However, where a powerful marketplace locks up a substantial portion of important developers or services, authorities may investigate whether rivals are being denied sufficient scale to compete effectively.

9. Interoperability Restrictions

AI agents frequently need to communicate with:

• other agents;
• databases;
• operating systems;
• websites;
• payment providers;
• cloud services;
• enterprise software.

A platform controlling these interfaces could potentially restrict rival agents.

Examples could include refusing API access, deliberately reducing functionality, imposing incompatible technical requirements, or allowing only the platform's own agents to use important system functions.

Interoperability therefore may become a major competition issue.

10. Tying and Bundling

Consider a company offering:

• a dominant operating system;
• an AI assistant;
• an agent marketplace;
• cloud services.

It could bundle these services.

For example:

Operating System + Default AI Assistant + Agent Marketplace

Competition concerns could arise if customers effectively have to accept the marketplace to obtain another important product or if bundling substantially restricts opportunities for competing marketplaces.

Bundling itself is common and often efficient. Competition law normally becomes concerned when market power and exclusionary effects are present.

11. Marketplace Fees

Marketplace operators may charge developers:

• listing fees;
• subscription fees;
• transaction commissions;
• API fees;
• payment-processing fees;
• advertising charges.

High fees alone do not automatically establish an antitrust violation.

Authorities would examine factors such as market power, alternatives available to developers, contractual restrictions and whether developers can communicate alternative purchasing methods to customers.

12. Most-Favoured-Nation Clauses

A marketplace could impose a clause requiring developers not to offer their agent more cheaply elsewhere.

For example:

Marketplace A price: $10
Marketplace B price: $8

If Marketplace A requires the developer to increase Marketplace B's price to at least $10, price competition between marketplaces may be reduced.

Such clauses are often called most-favoured-nation (MFN) or parity clauses.

Their legality depends heavily on market conditions and jurisdiction.

13. Algorithmic Ranking

AI marketplaces introduce another difficulty: the platform itself may decide which agent performs a task.

Instead of users browsing a traditional store, a general AI assistant might automatically choose:

Agent A rather than Agent B.

Consequently, algorithmic selection becomes the new shelf space.

Competition authorities may investigate whether selection criteria are genuinely based on quality, security, price or relevance, or whether they systematically favour affiliated services.

Important Case Laws and Precedents

14. United States v. Microsoft Corp.

United States Court of Appeals for the D.C. Circuit, 2001

Microsoft remains one of the most important precedents for competition in technology ecosystems.

Microsoft possessed monopoly power in PC operating systems. The litigation concerned conduct directed at technologies, particularly browsers and middleware, capable of weakening Microsoft's operating-system position.

The court found important aspects of Microsoft's exclusionary conduct unlawful under Section 2 of the Sherman Act.

Relevance to AI Agent Marketplaces

AI agents could become a new form of middleware.

An operating-system or platform provider could potentially disadvantage independent agents because those agents could reduce users' dependence on the platform's own interface.

The Microsoft precedent demonstrates that a dominant technology company cannot automatically use exclusionary practices merely because it owns the underlying platform.

The U.S. Department of Justice has itself invoked Microsoft when discussing contemporary platform competition.

15. Google Shopping

Google Search (Shopping), European Commission, 2017; General Court, 2021

The European Commission found that Google had favoured its own comparison-shopping service within general search results while disadvantaging competing comparison-shopping services.

The General Court largely upheld the Commission's decision.

Relevance

Imagine an AI assistant controlling access to an agent marketplace.

When asked:

“Find the best travel-planning agent.”

the platform consistently recommends its own travel agent ahead of independent competitors.

The Google Shopping litigation illustrates how preferential treatment by a powerful intermediary can become a competition concern where access to that intermediary materially affects rivals.

16. Google Android

European Commission, 2018; General Court, 2022

The Android litigation concerned restrictions connected with Google's Android ecosystem, including arrangements involving Google Search and Chrome.

The European Commission concluded that several practices unlawfully strengthened Google's search position. The General Court substantially upheld the decision while modifying aspects of the penalty.

AI Marketplace Relevance

The closest AI analogy would involve:

Operating system → AI assistant → Agent marketplace → Proprietary agents

Competition questions could arise where access to one important component is conditioned on installation or preferential treatment of another.

The case therefore provides an important framework for analysing defaults, pre-installation and ecosystem leverage.

17. Epic Games, Inc. v. Apple Inc.

United States Court of Appeals for the Ninth Circuit, 2023

The dispute between Epic and Apple concerned Apple's control over iOS app distribution and payment mechanisms.

Epic challenged Apple's restrictions under U.S. antitrust law and California unfair-competition law.

Apple prevailed on Epic's federal antitrust claims, while litigation concerning Apple's anti-steering restrictions produced significant consequences under California law.

AI Marketplace Relevance

AI-agent marketplaces may eventually raise similar disputes concerning:

• mandatory payment systems;
• developer commissions;
• marketplace access;
• alternative distribution;
• communication with customers.

The case also demonstrates an important principle: strong platform control does not automatically equal unlawful monopolization. Precise market definition and proof of anticompetitive effects remain critical.

18. Epic Games, Inc. v. Google LLC

U.S. District Court, Northern District of California

Epic separately challenged Google's practices concerning Android application distribution and payments.

A jury found for Epic on its antitrust claims in 2023, and the litigation subsequently produced remedies directed at competition in Android app distribution and billing.

AI Marketplace Relevance

The dispute is particularly relevant to future questions concerning whether an AI ecosystem can restrict:

• competing agent stores;
• alternative billing;
• developer distribution;
• direct relationships between developers and users.

It shows why marketplace competition itself can become an antitrust issue rather than competition merely between individual applications.

19. United States v. Google LLC — Search Monopolization

In August 2024, the U.S. District Court for the District of Columbia concluded that Google had unlawfully maintained monopolies in general search services and general search text advertising.

The case focused heavily on distribution agreements that made Google the default search engine across important access points.

Subsequent remedies included restrictions and requirements concerning distribution and access to certain search-related data and services.

AI Marketplace Relevance

AI agents may become a major gateway through which consumers reach online businesses.

If a dominant company secures exclusive or default placement for its AI assistant across smartphones, browsers or operating systems, authorities may examine whether those arrangements foreclose competing agents.

This precedent therefore has particularly strong relevance to default AI-agent distribution.

20. United States and Plaintiff States v. Apple Inc.

The U.S. Department of Justice and a group of states filed a monopolization case against Apple in 2024.

The government alleges that Apple uses contractual restrictions and control over important access points to protect its smartphone position and disadvantage technologies that could reduce dependence on the iPhone ecosystem.

This remains important to distinguish from final adjudicated precedent: the government's allegations are not themselves equivalent to a final finding of liability.

AI Marketplace Relevance

The theory is nevertheless highly relevant to agent markets because independent AI agents may eventually function across multiple devices and services.

A platform could potentially protect its ecosystem by restricting:

• cross-platform agents;
• alternative agent marketplaces;
• system-level API access;
• authentication capabilities;
• interoperability.

Those are conceptually similar to the types of platform-access questions raised in the Apple litigation.

21. Additional Competition Risks

Several other practices could attract scrutiny.

Predatory strategies: A marketplace might subsidize its own agent services heavily to gain scale and later attempt to recoup losses after rivals exit. Predatory-pricing rules, however, usually impose demanding legal requirements.

Discriminatory API access: Affiliated agents might receive faster or more capable APIs than independent competitors.

Data foreclosure: Competitors could be denied information necessary for training, personalization or effective agent performance.

Acquisitions: A dominant marketplace could repeatedly acquire emerging agent developers before they become substantial competitive threats.

Multi-homing restrictions: Developers might be prevented from listing agents on competing marketplaces.

Steering restrictions: Developers might be prevented from telling users that the same service is available through another channel.

22. Competition Benefits of AI Agent Marketplaces

Competition law should also recognize substantial efficiencies.

Agent marketplaces could:

• reduce software-discovery costs;
• give small developers access to large customer bases;
• provide common security standards;
• simplify authentication;
• reduce transaction costs;
• facilitate interoperability;
• encourage specialized agent development;
• improve consumer comparison.

Therefore, neither vertical integration nor marketplace operation is inherently anticompetitive.

The central legal question is generally whether particular conduct protects competition on the merits or improperly restricts the competitive process.

23. Emerging Regulatory Attention

The subject is no longer merely theoretical.

In July 2026, France's competition authority published an opinion specifically examining competitive conditions in the AI-agent sector after consulting industry and institutional stakeholders.

In the United States, proposed legislation released in June 2026 has also focused on the possibility that dominant technology platforms could restrict competing AI agents, steer consumers toward affiliated products or control access to important digital services.

These developments indicate that competition policy is beginning to treat agents as potential digital intermediaries and gatekeepers, rather than simply another category of software.

24. Practical Competition-Law Framework

A competition authority examining an AI-agent marketplace could proceed through the following questions:

First, what is the relevant market?

Second, does the marketplace operator possess substantial or dominant market power?

Third, what barriers prevent competing marketplaces from expanding?

Fourth, does the operator compete against businesses that depend upon its marketplace?

Fifth, does it favour its own agents through ranking, defaults, data or technical access?

Sixth, are developers restricted from multi-homing or using competing payment systems?

Seventh, are important APIs or interoperability mechanisms being withheld?

Eighth, do exclusive contracts foreclose meaningful distribution channels?

Ninth, are there legitimate security, privacy, quality or efficiency justifications?

Finally, do the restrictions produce competitive harm that outweighs legally relevant efficiencies under the applicable jurisdiction's rules?

Conclusion

AI agent marketplaces could become a significant layer of the digital economy because they may control not merely where software is distributed but which software an AI chooses and uses on behalf of consumers.

This creates several potential competition-law concerns: gatekeeper power, self-preferencing, defaults, exclusive distribution, tying, discriminatory API access, interoperability restrictions, data advantages, steering restrictions and ecosystem lock-in.

Existing authorities such as United States v. Microsoft, Google Shopping, Google Android, Epic Games v. Apple, Epic Games v. Google, United States v. Google, and the ongoing U.S. v. Apple litigation provide important frameworks for analysing these issues.

The distinctive feature of the emerging agent economy is that the intermediary may increasingly make the selection decision itself. As a result, control over agent discovery, ranking, authorization, interoperability and execution may become as important to future competition policy as control over app stores and search distribution has been in earlier generations of digital markets.

Add a jurisdiction-by-jurisdiction legal framework

Add a jurisdiction-by-jurisdiction legal framework

Clarify the difference between cases and allegations

Add a table linking cases to agent risks