AI Agent Marketplaces and Autonomous Service Competition Risks

1. Meaning

AI agent marketplaces are digital platforms where users, businesses, or other software agents can discover, compare, purchase, subscribe to, or deploy AI agents capable of performing tasks autonomously.

An AI agent may be able to:

search for information;

negotiate with suppliers;

purchase products;

book services;

execute financial or business instructions;

write and deploy software;

manage advertising;

interact with other agents;

select another AI service;

continuously optimize decisions.

An AI agent marketplace can therefore become an intermediary between:

Users → Marketplace → AI agents → External services

or even:

Agent A → Marketplace → Agent B → External service.

The competition-law significance is that the marketplace operator may control discovery, ranking, access, data, APIs, payments, identity, reputation and transaction execution simultaneously.

2. Why AI Agent Marketplaces Are Different

Traditional digital marketplaces generally connect humans with businesses.

AI-agent marketplaces may connect:

humans + autonomous agents + businesses + other agents.

This changes competitive dynamics because an agent can potentially make purchasing or switching decisions much faster and at much larger scale than a human.

For example, one AI agent might automatically compare:

500 insurance offers;

1,000 cloud services;

hundreds of software products;

thousands of suppliers.

Consequently, control over the agent-discovery layer could become commercially significant.

3. Basic Market Structure

A future AI-agent ecosystem could contain:

Layer 1 — Foundation models

Large language models and multimodal models.

Layer 2 — Agent frameworks

Systems allowing models to perform autonomous tasks.

Layer 3 — Agent marketplace

Platform where agents are discovered and deployed.

Layer 4 — Tools and APIs

Search, payment, communication, databases and other capabilities.

Layer 5 — Autonomous services

Agents performing:

shopping;

travel;

accounting;

legal research;

marketing;

coding;

logistics;

procurement.

Layer 6 — End users

Individuals and businesses using autonomous services.

Competition problems can arise when one company controls multiple layers.

4. Marketplace Gatekeeper Risk

An AI-agent marketplace may become a gateway through which users discover competing agents.

The platform could potentially control:

search ranking;

recommendations;

default agents;

visibility;

reviews;

pricing;

API access;

transaction fees;

certification.

If a marketplace operator also operates its own competing AI agent, a conflict may arise.

For example:

Marketplace X hosts 1,000 third-party agents but also owns Agent X.

X might theoretically give its own agent:

preferred ranking;

better API access;

lower commissions;

greater visibility;

default status.

Whether such conduct infringes competition law would depend on market power, foreclosure effects and other legal factors.

5. Self-Preferencing

Self-preferencing occurs when a platform gives preferential treatment to its own downstream service compared with competing services.

In an AI-agent marketplace, possible examples include:

placing its own agent first;

recommending its own agent by default;

giving its own agent privileged API access;

reducing its own transaction fees;

providing third-party agents with slower infrastructure;

restricting competitors' access to marketplace data.

The competition concern becomes stronger where the marketplace has substantial market power and the conduct materially disadvantages competing agents.

6. Ranking Manipulation

Ranking may become especially important because autonomous agents may not browse dozens of alternatives themselves.

Instead, an agent could receive:

"Choose the best accounting service."

The marketplace algorithm may determine which services are presented.

If ranking is manipulated, the marketplace could influence autonomous purchasing decisions at enormous scale.

Potential issues include:

biased ranking;

paid placement;

discriminatory algorithms;

hidden preferential treatment;

exclusion from recommendation systems.

7. Default-Agent Competition

A platform could designate its own AI agent as the default.

For example:

Device → operating system → default assistant → marketplace → service selection.

Defaults can influence user behavior because users may not change them.

The competition question is whether the default arrangement merely improves convenience or whether it forecloses rival agents.

8. Agent Lock-In

AI agents may develop persistent:

user preferences;

transaction histories;

credentials;

workflows;

memories;

API connections;

reputation profiles.

Switching agents may therefore become costly.

For example:

Agent A has learned a company's procurement preferences over three years.

Moving to Agent B might require transferring:

historical data;

instructions;

preferences;

credentials;

workflows.

This creates potential agent portability and switching-cost concerns.

9. Interoperability

Interoperability is particularly important because agents may need to communicate with:

marketplaces;

payment systems;

cloud platforms;

databases;

other agents.

If a dominant platform prevents rival agents from interoperating effectively, competitors may face substantial barriers.

Potential restrictions include:

closed APIs;

incompatible protocols;

access fees;

authentication restrictions;

technical throttling.

10. Data Advantage

AI marketplaces can collect enormous amounts of information about:

user requests;

purchases;

agent performance;

conversion rates;

prices;

suppliers;

user preferences;

agent failures.

The marketplace operator may then use this information to improve its own competing agent.

This creates a potential:

marketplace information → competing agent → better performance → increased marketplace dominance

feedback loop.

11. Competitor Data Use

Suppose independent agents sell services through Platform P.

P learns:

which services sell most;

which customers pay most;

which prices convert best;

which agent functions are most popular.

P then launches a competing agent based on those insights.

The competition question is whether this represents legitimate innovation or whether the platform is exploiting competitively sensitive information obtained through its intermediary position.

12. Tying and Bundling

An AI ecosystem might combine:

operating system + AI assistant + agent marketplace + cloud services + payment system.

A platform could require users to use its own agent as a condition for accessing another service.

Potential theories include:

tying;

bundling;

leveraging;

exclusionary conduct.

The legal analysis depends on market definition and whether the arrangement forecloses competitors.

13. Exclusive Dealing

A marketplace might offer developers:

"Lower commission if you agree not to list your agent on competing marketplaces."

Exclusive arrangements are not automatically unlawful.

Relevant considerations include:

duration;

coverage;

market power;

availability of alternatives;

foreclosure;

switching costs.

14. Commission and Fee Discrimination

An AI marketplace may charge:

5% commission to its own agents;

20% to independent agents.

If the marketplace has market power, discriminatory commercial terms could become relevant to competition-law analysis.

The same issue could arise with:

API fees;

computing resources;

transaction charges;

advertising fees;

certification fees.

15. Autonomous Purchasing and Algorithmic Coordination

One of the most novel risks concerns agent-to-agent interaction.

Suppose thousands of autonomous purchasing agents interact with thousands of seller agents.

The agents may:

observe prices;

react to competitors;

change prices automatically;

negotiate;

optimize margins.

This could produce highly rapid market responses.

Competition authorities may need to distinguish between:

Independent algorithmic optimization

Each firm independently programs its system.

and

Coordinated conduct

Algorithms are designed or used to implement an agreement or concerted practice.

The fact that algorithms communicate or produce parallel prices does not, by itself, establish unlawful coordination.

16. Algorithmic Collusion

AI agents may potentially facilitate:

price coordination;

market allocation;

output coordination;

information exchange.

For example:

Seller Agent A communicates with Seller Agent B and agrees on minimum prices.

That would raise conventional cartel concerns even though an AI system executed the agreement.

A harder question arises where algorithms independently learn similar strategies without explicit communication.

That issue remains legally and economically complex.

17. Autonomous Negotiation

AI agents may negotiate contracts automatically.

For example:

Buyer Agent → Seller Agent → price negotiation → delivery terms → contract.

Competition questions could arise if agents collectively facilitate:

uniform pricing;

coordinated purchasing;

exclusion of particular suppliers;

discriminatory treatment.

The relevant competition law would focus on the underlying conduct rather than simply the fact that AI performed it.

18. Agent Marketplace Network Effects

A marketplace can benefit from:

more users → more developers → more agents → greater usefulness → more users.

This creates network effects.

A successful platform may therefore become difficult for a rival marketplace to challenge.

Potential competitive concerns include:

exclusivity;

interoperability restrictions;

data portability restrictions;

acquisition of emerging rivals;

self-preferencing.

19. Killer Acquisitions

Large AI companies could potentially acquire:

successful agent developers;

specialized autonomous-service providers;

agent orchestration companies;

interoperability tools.

A target might have relatively low current revenue but possess strategically important technology or users.

Merger analysis may therefore focus on:

potential competition;

innovation;

technology;

data;

ecosystem effects.

20. Refusal to List an Agent

Suppose a dominant marketplace refuses to list a rival agent.

The competition analysis may involve:

dominance;

importance of the marketplace;

indispensability;

availability of alternatives;

justification for refusal;

effect on competition.

A platform is not automatically required to host every competing service.

21. Interoperability as a Competitive Remedy

Competition authorities could potentially consider remedies such as:

API access;

data portability;

interoperability;

non-discrimination;

transparent ranking;

restrictions on self-preferencing;

separation of marketplace and downstream services.

The appropriate remedy would depend on the specific competitive harm.

22. Important Case Laws

Because autonomous AI-agent marketplaces are a relatively new phenomenon, there are not yet many reported cases directly concerning AI-agent marketplaces. Existing platform, interoperability, tying, refusal-to-deal and algorithmic-competition authorities provide the relevant legal foundations.

1. United States v. Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001)

Microsoft used its operating-system dominance in ways that affected competition from web browsers.

Principle

A dominant platform can face antitrust scrutiny when it uses control over a platform to restrict or disadvantage competing products.

AI-agent relevance

The case is relevant to:

default agents;

operating-system integration;

API restrictions;

platform control;

interoperability.

23. Google Shopping, European Commission, Case AT.39740

The European Commission found Google had abused its dominant position by favoring its comparison-shopping service in general search results.

Principle

A dominant platform's preferential treatment of its own downstream service can raise competition concerns.

AI-agent relevance

An AI-agent marketplace could face analogous scrutiny if it systematically favors its own agent over competing agents.

24. Google Android, European Commission, Case AT.40099

The European Commission examined contractual restrictions involving Android and Google's related services.

Principle

Tying and ecosystem restrictions can potentially reinforce dominance in connected markets.

AI-agent relevance

The case is useful for analyzing:

operating system → default AI assistant → agent marketplace → downstream services.

25. Bronner v Mediaprint, Case C-7/97

The Court of Justice considered refusal to provide access to an infrastructure controlled by a dominant undertaking.

Principle

Dominance alone does not automatically create a general obligation to supply competitors. The strict conditions surrounding refusal-to-deal doctrine must be examined.

AI-agent relevance

This is relevant when an AI marketplace refuses:

API access;

marketplace access;

infrastructure access;

interoperability.

26. IMS Health v NDC Health, Case C-418/01

The case concerned access to a commercially valuable information structure.

Principle

Under exceptional circumstances, refusal to license or provide access to an indispensable resource can constitute abuse of dominance.

AI-agent relevance

Potentially relevant to:

proprietary agent datasets;

interoperability;

agent APIs;

specialized AI information systems.

27. Slovak Telekom v Commission, Joined Cases C-152/19 P and C-165/19 P

The case concerned access to telecommunications infrastructure and exclusionary effects.

Principle

Access conditions imposed by a vertically integrated dominant undertaking can be scrutinized when they restrict downstream competition.

AI-agent relevance

It provides an analytical framework for:

dominant AI infrastructure → restrictive access → downstream agent competition.

28. Intel v Commission, Case C-413/14 P

The Court of Justice examined exclusivity-related rebates offered by a dominant undertaking.

Principle

The circumstances and actual or potential exclusionary effects of allegedly anticompetitive rebates may require detailed economic examination.

AI-agent relevance

This can inform analysis of:

marketplace discounts;

developer incentives;

exclusivity payments;

preferential commissions.

29. Comcast Corp. v. Behrend, 569 U.S. 27 (2013)

The U.S. Supreme Court considered whether a proposed damages methodology corresponded to the theory of antitrust liability.

Principle

Economic damages methodology must be connected to the alleged anticompetitive conduct.

AI-agent relevance

If an AI marketplace is alleged to have harmed thousands of developers or users, proving aggregate competitive injury may require a methodology connected to the specific theory of harm.

30. United States v. Apple Inc., 791 F.3d 290 (2d Cir. 2015)

The case concerned Apple's involvement in alleged coordination among book publishers.

Principle

Technology platforms can be subject to ordinary antitrust rules where their conduct facilitates coordination among market participants.

AI-agent relevance

The case is particularly useful for considering whether an AI platform facilitates coordination among autonomous seller or buyer agents.

31. Potential Theories of Harm

32. Economic Effects

The relevant effects may include:

Reduced innovation

Rivals may find it harder to enter.

Higher prices

Marketplace commissions or downstream service prices may rise.

Reduced choice

Users may see fewer agents.

Lower quality

Artificial restrictions may reduce competitive pressure.

Slower innovation

Developers may have less incentive to create alternative agents.

Increased switching costs

Users and businesses may become dependent on one ecosystem.

However, integration can also produce legitimate efficiencies such as:

better security;

reduced transaction costs;

improved agent reliability;

fraud prevention;

easier discovery;

better interoperability;

improved user experience.

33. Special Problem: Who Is the Decision-Maker?

Traditional competition law generally assumes that human firms make economic decisions.

Autonomous-agent markets complicate this assumption.

An agent may:

choose suppliers;

negotiate prices;

change purchasing strategies;

select another agent;

execute transactions.

The legal responsibility nevertheless ultimately attaches to the relevant human or corporate actors under applicable competition law.

The use of AI does not itself eliminate responsibility for unlawful agreements or exclusionary conduct.

34. Future Regulatory Issues

Important future questions include:

Should AI agents be portable between marketplaces?

Should agents have standardized APIs?

Can marketplace operators use third-party agent performance data?

Should ranking algorithms be independently auditable?

Should dominant AI platforms be subject to interoperability obligations?

How should algorithmic coordination be detected?

Should mergers involving strategically important agents receive enhanced scrutiny?

How should autonomous purchasing affect competition enforcement?

Who controls an agent's accumulated user data?

How should competition authorities distinguish innovation from exclusion?

35. Exam-Oriented Framework

For an examination problem involving an AI-agent marketplace, use this sequence:

Step 1 — Identify the market

Determine whether the relevant market concerns:

AI agents;

agent marketplaces;

AI infrastructure;

a particular autonomous service;

or a connected ecosystem.

Step 2 — Determine market power

Examine:

market share;

network effects;

switching costs;

data advantages;

entry barriers;

interoperability.

Step 3 — Identify the conduct

Ask whether there is:

self-preferencing;

tying;

exclusive dealing;

refusal to deal;

discrimination;

data exploitation;

interoperability restriction;

algorithmic coordination.

Step 4 — Examine effects

Consider:

foreclosure;

innovation;

price;

quality;

consumer choice;

entry.

Step 5 — Examine efficiencies

Consider whether the conduct improves:

security;

privacy;

reliability;

innovation;

functionality.

Step 6 — Consider remedies

Potential remedies include:

interoperability;

data portability;

non-discrimination;

transparent ranking;

access obligations;

behavioral restrictions;

structural remedies where legally justified.

36. Conclusion

AI-agent marketplaces could become important competitive bottlenecks because they may control discovery, ranking, data, APIs, payments, identity and access to autonomous services simultaneously.

The principal competition issues are:

self-preferencing;

default-agent restrictions;

data advantages;

agent lock-in;

interoperability restrictions;

exclusive dealing;

tying and bundling;

algorithmic coordination;

discriminatory marketplace access;

AI-agent acquisitions and ecosystem consolidation.

The principal traditional authorities—Microsoft, Google Shopping, Google Android, Bronner, IMS Health, Slovak Telekom, Intel, and Apple—provide useful legal frameworks, but the application of those principles to genuinely autonomous AI-agent ecosystems will depend heavily on the precise market structure, degree of autonomy, platform power, technical architecture and competitive effects.