Competition Law and AI Agent Marketplace Competition

1. Introduction

An AI agent marketplace is a digital platform through which businesses or consumers can discover, obtain, deploy, integrate, or purchase AI agents. These agents may perform tasks such as customer support, research, coding, workflow automation, purchasing, data analysis, scheduling, or interaction with other software.

From a competition-law perspective, these marketplaces are important because the marketplace operator may simultaneously control several layers of the AI ecosystem: the marketplace itself, the underlying foundation model, cloud infrastructure, identity and payment systems, user data, ranking mechanisms, and sometimes competing first-party AI agents.

As of 2026, competition authorities are examining agentic AI specifically. The French Competition Authority has identified AI-agent distribution channels and model-as-a-service platforms as areas requiring particular attention, including questions concerning access, visibility, ranking, default agents and partnerships between major digital firms and AI-agent developers.

There is still limited decided case law dealing specifically with AI-agent marketplaces. Therefore, the most important legal precedents come from app stores, search engines, operating systems, online marketplaces and other digital platforms. Those principles can potentially be applied to AI-agent marketplaces.

2. Structure of an AI Agent Marketplace

A typical marketplace can involve several interconnected layers:

AI infrastructure → foundation model → agent-development framework → marketplace → AI agents → complementary services → end users.

Competition may occur at every layer. For example, one company might supply a foundation model while simultaneously operating the marketplace through which competing agents using other models must reach customers.

This creates what competition lawyers often describe as vertical integration.

Vertical integration is not inherently unlawful. It may reduce costs, improve security and create better-integrated products. Competition concerns arise where market power at one level is used to disadvantage competitors at another level.

3. Market Definition

Before establishing dominance or monopolization, authorities generally need to understand the relevant competitive market.

Potential AI-agent markets could include:

• AI-agent marketplaces;
• enterprise-agent distribution platforms;
• consumer AI assistants;
• specialised AI agents;
• foundation-model services;
• agent-development platforms;
• agent hosting or inference services;
• cloud infrastructure for AI agents; and
• particular categories such as coding, commerce or productivity agents.

An important question is whether different marketplaces are sufficiently interchangeable.

For example, an enterprise marketplace designed around corporate software integration might not provide an effective substitute for a consumer-oriented agent store.

Authorities would examine factors including switching behaviour, prices, technical compatibility, functionality, developer migration costs and user preferences.

4. Network Effects

Agent marketplaces may exhibit strong network effects.

More users attract more agent developers because developers obtain a larger potential customer base.

More developers create more agents.

More agents make the marketplace more attractive to users.

This produces a reinforcing cycle:

Users → developers → more agents → better marketplace → more users.

Network effects themselves are not anticompetitive. However, when combined with high switching costs, proprietary interfaces and large amounts of data, they can make entry considerably more difficult.

5. Default AI Agents

One significant competition issue concerns default placement.

Imagine that a major operating system provides its own AI agent automatically while competing agents require several additional installation or permission steps.

Even when users technically remain free to choose another agent, default placement can materially influence distribution.

The French Competition Authority's 2026 AI-agent assessment specifically highlighted the effective ability of users to choose competing agents instead of agents integrated by default into major digital ecosystems.

The competition question therefore becomes whether the default arrangement represents ordinary product integration or exclusionary leveraging of market power.

6. Self-Preferencing

A marketplace operator could also develop its own agents.

Suppose an agent store contains:

• Marketplace Agent A — owned by the platform;
• Independent Agent B;
• Independent Agent C.

The operator could potentially influence competition through search rankings, recommendation algorithms, featured placements, certification systems or default selections.

For example, its ranking algorithm might systematically display Agent A first even where B or C would otherwise compete effectively.

Such behaviour can raise self-preferencing concerns.

The issue is especially significant because AI agents themselves may become intermediaries deciding which businesses or services consumers see.

The French authority has therefore highlighted the parameters controlling the selection, ranking and recommendation of services by AI agents as a competition concern.

7. Marketplace Data Advantage

Marketplace operators may obtain substantial information about independent developers, including:

• agent usage;
• customer demand;
• conversion rates;
• prices;
• retention;
• popular features;
• transaction volumes;
• customer categories; and
• performance information.

This can create a dual-role problem.

The platform provides marketplace infrastructure while simultaneously competing against businesses using that infrastructure.

Competition concerns become stronger if confidential marketplace information is used to identify successful independent agents and develop competing first-party alternatives.

The European Commission's Amazon Marketplace proceedings are particularly relevant by analogy. Amazon ultimately committed not to use certain non-public marketplace seller data for its competing retail operations.

8. Interoperability and Portability

Another important issue is whether developers can move their agents between ecosystems.

Suppose Agent X works only with:

Marketplace X + Model X + Cloud X + API X.

Moving it to Marketplace Y could require substantial redevelopment.

Such technical dependence can increase switching costs and strengthen ecosystem lock-in.

Competition authorities could consequently examine restrictions involving:

• APIs;
• agent protocols;
• authentication;
• memory formats;
• tool interfaces;
• model compatibility;
• data portability; and
• third-party integrations.

Interoperability becomes especially important when access to a dominant ecosystem is necessary for effective competition.

9. Tying and Bundling

AI ecosystems create numerous possibilities for tying.

For example:

Cloud service + AI model + agent marketplace

or:

Operating system + built-in AI agent

or:

Productivity suite + proprietary AI agent

Bundling can create substantial efficiencies. Consumers may obtain an integrated product without separately installing several services.

But competition concerns can arise where a dominant supplier effectively requires customers to obtain another product or makes competing products significantly harder to use.

The classic Microsoft litigation remains highly relevant because it addressed both interoperability and the bundling of Windows Media Player with Windows.

10. Exclusivity Agreements

An AI marketplace might offer developers favourable conditions if they distribute their agents exclusively through that marketplace.

Possible arrangements include:

"Receive reduced marketplace fees if your agent is unavailable elsewhere."

Such arrangements are not automatically unlawful.

Their competitive effect depends on factors such as market power, duration, market coverage, available alternative distribution channels and whether competitors can realistically reach developers or customers.

If a powerful marketplace locks up many leading agents, competing marketplaces could struggle to attract users.

11. Most-Favoured-Nation Clauses

An agent marketplace might require developers not to offer their agents elsewhere at lower prices or on better terms.

For example:

Agent Developer X must not offer its agent cheaper through another marketplace.

Such MFN/parity clauses can sometimes reduce transaction costs or prevent free riding, but broad clauses can also reduce incentives for competing marketplaces to offer developers lower commissions.

A new marketplace may find it difficult to attract developers through lower fees if developers cannot pass those savings to customers.

12. Marketplace Fees

Marketplace operators may charge:

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

High fees alone do not necessarily constitute an antitrust violation.

Competition-law scrutiny becomes more likely where market power combines with exclusionary conditions—for example, restrictions preventing developers from directing customers toward alternative purchasing channels.

App-store cases provide an important analogy.

Important Case Laws and Competition Proceedings

Because dedicated AI-agent-marketplace jurisprudence remains developing, the following precedents provide the strongest analogies.

1. Google and Alphabet v Commission — Google Shopping

Case C-48/22 P, Court of Justice, 2024

Google operated general search while also supplying its own comparison-shopping service.

The European Commission concluded that Google had favoured its own comparison-shopping service over competing services. After proceedings before the General Court, the Court of Justice dismissed Google's appeal in September 2024 and upheld the judgment concerning the Commission's €2.4 billion fine.

Relevance to AI Agent Marketplaces

The case is particularly important for self-preferencing and leveraging.

Imagine a dominant AI-agent marketplace displaying:

Platform-owned agent — Position 1

while comparable independent agents consistently receive inferior visibility.

The Google Shopping judgment provides an important analytical precedent for considering whether the platform is using control over an important distribution mechanism to favour its adjacent service.

However, it does not establish that every instance of first-party preference is automatically unlawful; the surrounding market circumstances and competitive effects remain important.

2. Google Android

Case AT.40099; Google and Alphabet v Commission, T-604/18

The European Commission's Android decision concerned Google's contractual practices involving Android, the Play Store, Google Search and Chrome. The Commission treated certain product bundling, exclusivity and anti-fragmentation practices as an infringement of Article 102 TFEU.

The General Court's 2022 judgment addressed the Android ecosystem as a multi-sided platform environment and substantially upheld the Commission's findings, while adjusting aspects of the decision and fine.

Relevance

The analogy to agent marketplaces is strong.

A company could potentially control:

AI operating environment + agent marketplace + foundation model + proprietary agent.

Competition concerns could arise if marketplace access is conditioned on installation, promotion or preferential treatment of the operator's own agent or model.

The case therefore illustrates how competition law can examine an entire digital ecosystem rather than treating every component completely independently.

3. Microsoft v Commission

Case T-201/04

Microsoft concerned Microsoft's dominant Windows operating-system position and practices involving interoperability information and Windows Media Player.

The Court of First Instance upheld central parts of the Commission's findings concerning Microsoft's refusal to supply interoperability information and its tying of Windows Media Player with Windows.

Relevance

The decision provides two particularly important lessons for agent ecosystems.

First, interoperability can become competitively significant where rivals need access to technical information to compete effectively.

Second, bundling a complementary product with a powerful platform can create foreclosure concerns.

The analogy would be particularly relevant if an AI platform bundled its proprietary agent in circumstances that materially restricted rival agents' ability to compete.

4. Amazon Marketplace

Case AT.40462

The European Commission investigated Amazon's position as both marketplace operator and retailer.

The concern involved Amazon's access to non-public information generated through independent sellers' marketplace activities.

Amazon ultimately committed not to use specified non-public third-party seller information for its competing retail operations.

Relevance

This is one of the closest analogies to an AI-agent marketplace.

An AI marketplace operator could observe:

which independent agent is growing → which features customers prefer → which prices work → which customers use the agent.

If the marketplace then uses privileged non-public information to launch a competing first-party agent, competition authorities could examine whether its dual role creates an unfair competitive advantage.

5. Amazon Buy Box

Case AT.40703

The European Commission expressed preliminary concerns about criteria governing Amazon's prominent Buy Box and Prime eligibility.

Amazon subsequently committed to apply non-discriminatory conditions and criteria for identifying the Featured Offer and made related commitments concerning competing offers and Prime eligibility.

Relevance

The analogy to AI-agent ranking systems is particularly direct.

In an AI-agent marketplace, the equivalent of the Buy Box might be:

"Recommended Agent"

or

"Best Agent for this task."

If most consumers simply select the recommended agent, control over that recommendation becomes commercially powerful.

Competition authorities could therefore examine whether ranking criteria unfairly favour the marketplace operator's agents.

6. Apple — App Store Practices (Music Streaming)

Case AT.40437

This European Commission proceeding concerned restrictions imposed by Apple on music-streaming application developers.

In March 2024, the Commission concluded that Apple's anti-steering provisions prevented music-streaming developers from adequately informing iOS users about alternative and potentially cheaper subscription possibilities outside the app. The decision was adopted under Article 102 TFEU.

Relevance

Similar restrictions could emerge in agent marketplaces.

For example, a marketplace might prohibit an agent developer from telling customers:

"You can subscribe directly from our website."

Competition authorities could examine whether such anti-steering restrictions prevent developers from competing effectively over price and distribution.

7. Apple DMA Anti-Steering Decision — 2025

Although this is a Digital Markets Act enforcement decision rather than a traditional Article 102 antitrust judgment, it is highly relevant to future AI marketplace regulation.

In April 2025, the European Commission found Apple in breach of the DMA's anti-steering obligation and imposed a €500 million fine. The Commission stated that developers should be able to inform customers about alternative offers outside Apple's App Store and steer customers toward them.

Relevance

If major AI-agent marketplaces eventually operate as significant digital gateways, similar regulatory principles concerning alternative distribution and developer steering may become relevant.

The French Competition Authority has already suggested considering whether marketplaces distributing AI models should potentially fall within the DMA's core-platform-service framework.

13. AI Agents as Gatekeepers Themselves

There is an additional issue that distinguishes AI agents from traditional app stores.

An ordinary app store primarily determines which applications users discover.

An AI agent may go further and make the choice itself.

For example, a consumer could instruct:

"Find the cheapest suitable hotel and book it."

The agent might decide:

which search provider → which booking platform → which hotel → which payment provider.

The consumer may never see competing alternatives.

Consequently, the agent's recommendation architecture could become a new competitive gateway.

This makes transparency concerning commercial relationships, ranking parameters and conflicts of interest increasingly significant.

14. Agent-to-Agent Competition

Future marketplaces may involve agents negotiating directly with other agents.

A purchasing agent might communicate automatically with several seller agents and select an offer.

This could increase competition because agents can compare thousands of offers more efficiently than consumers.

However, autonomous pricing and negotiation also raise questions about algorithmic coordination.

The UK CMA has specifically identified possible competition risks where AI systems influence pricing and noted theoretical scenarios in which autonomous agents could converge toward outcomes resembling tacit coordination.

Existing competition rules concerning agreements and concerted practices would therefore remain important where legally sufficient coordination between undertakings can be established.

15. Killer Acquisitions and Strategic Investments

Large ecosystem operators may invest in or acquire promising AI-agent developers.

Such transactions are not inherently problematic. They can provide developers with capital, computing resources and distribution.

Authorities may nevertheless examine whether transactions:

• remove emerging competitors;
• consolidate important agent technology;
• give a platform influence over competing agents;
• restrict multi-homing;
• strengthen control over important distribution channels; or
• combine important datasets and infrastructure.

The French Competition Authority's 2026 opinion specifically recommends attention to investments and partnerships between major digital operators and competing AI-agent developers.

16. Multi-Homing and Switching Costs

A healthy marketplace structure is generally easier to maintain when developers and users can multi-home.

For developers, this means distributing an agent across several platforms.

For consumers, it means using multiple competing agents without excessive switching costs.

Potential barriers include proprietary agent formats, exclusive contracts, non-portable memory, incompatible APIs, technical restrictions and loss of accumulated data.

When these barriers become substantial, a marketplace's installed user base may become increasingly difficult for competitors to challenge.

17. Relevant Competition-Law Framework

In the European Union, the main traditional competition provisions potentially relevant to agent marketplaces are:

Article 101 TFEU — agreements and concerted practices restricting competition.

This could potentially cover anticompetitive exclusivity arrangements, coordination between marketplace participants or certain restrictive agreements.

Article 102 TFEU — abuse of a dominant position.

Potential theories include tying, discriminatory access, exclusionary self-preferencing, certain refusals involving interoperability, unfair conditions and other forms of exclusionary leveraging.

The Digital Markets Act may separately become increasingly significant where AI services or marketplaces fall within its scope. The European Commission's 2026 DMA review materials indicate continuing examination of whether AI services should be designated within existing categories such as virtual assistants or whether further investigation concerning AI services as potential core platform services is warranted.

18. Pro-Competitive Effects

Competition analysis must also account for substantial benefits.

Agent marketplaces can:

• lower software-development costs;
• allow small developers to reach customers;
• reduce search costs;
• improve interoperability;
• facilitate specialised agents;
• create new business models;
• increase price comparison;
• automate repetitive transactions; and
• reduce barriers to using sophisticated AI.

The presence of a powerful marketplace therefore does not by itself establish a competition-law violation.

The central issue is generally whether particular conduct protects competition through legitimate product improvement or instead uses market power to exclude competitive alternatives without sufficient justification.

19. Likely Future Enforcement Questions

Competition authorities examining an AI-agent marketplace are likely to ask several connected questions:

Market power: Does the marketplace function as an important gateway?

Distribution: Can rival agents realistically reach customers elsewhere?

Ranking: Are first-party and third-party agents treated according to defensible criteria?

Defaults: Can users easily replace the marketplace operator's default agent?

Data: Does the operator exploit confidential information generated by competing developers?

Interoperability: Can agents communicate with competing models, tools and platforms?

Multi-homing: Can developers distribute agents through multiple marketplaces?

Tying: Must customers purchase the operator's cloud, model or other services?

Steering: Can developers communicate alternative prices and distribution channels?

Acquisitions: Are promising independent agent developers being acquired or strategically controlled?

Algorithmic coordination: Are autonomous commercial agents facilitating prohibited coordination between businesses?

These questions broadly reflect the concerns now appearing in competition-authority work on agentic AI.

Conclusion

Competition law and AI agent marketplace competition is likely to become an important branch of digital-platform competition law. AI marketplaces combine characteristics previously examined separately in app stores, operating systems, online marketplaces, search engines and cloud ecosystems.

The six especially useful precedents are Google Shopping (C-48/22 P), Google Android (T-604/18), Microsoft v Commission (T-201/04), Amazon Marketplace (AT.40462), Amazon Buy Box (AT.40703), and Apple App Store Practices (AT.40437). They provide established legal frameworks for analysing self-preferencing, tying, interoperability restrictions, marketplace-data advantages, discriminatory ranking and anti-steering practices.

The distinctive feature of AI-agent markets is that the marketplace may not merely distribute software: AI agents themselves may increasingly select, rank, negotiate with and purchase from other businesses on users' behalf. Consequently, control over agent distribution, defaults, interoperability, data and recommendation architecture could become a significant source of market power. Current competition-authority work, particularly the French Competition Authority's July 2026 opinion, shows that these questions have already moved from theoretical discussion into active regulatory scrutiny.