Consumers and businesses have long relied on agents and intermediaries to work on their behalf, including to help reduce search and transaction costs, address information asymmetries and provide expertise or administrative support. Agentic AI systems have begun to extend the range and complexity of activities that can be delegated to computers. By combining capabilities such as planning, memory and tool use, they can carry out multi-step tasks and act on behalf of users, including as buyers and sellers, with varying degrees of human oversight (OECD, 2026[1]).1
Building on wider advances in AI, these systems could support productivity growth by reducing operational and co-ordination costs and allowing people to devote more time to supervisory, strategic and creative activities. They may also lower barriers to entry and reduce the scale and specialist resources needed to carry out some activities, supporting innovation and the development of new products and services (OECD, 2025[2]). However, realising these benefits will depend on systems’ capabilities, reliability and effective deployment. The technology continues to develop, and adoption remains early and uneven, with available surveys indicating growing use alongside cautious delegation and limited maturity in organisational governance (Stackoverflow, 2026[3]; Bayiates, 2026[4]; PWC, 2025[5]).
Agentic AI systems also share, and in some cases may amplify, wider risks associated with AI, including safety and cybersecurity concerns, privacy breaches, bias, manipulation and misinformation. Greater delegation may amplify the consequences of errors, misuse or misalignment, including through unauthorised actions and financial commitments. Limited visibility into how systems select information, use tools and reach outputs or decisions may complicate oversight, accountability and redress. Interactions between increasingly autonomous systems may create additional risks that are difficult to anticipate, particularly where actions compound across agents, firms or markets (Hammond et al., 2025[6]).
These potential benefits and risks raise broader economic and societal questions that extend beyond competition policy.2 Competition policy, however, remains an important part of the wider policy response, helping to foster competitive markets and support innovation. This paper focusses on two dimensions: competition in the supply of agentic AI systems and the effects of their use on competition in other markets.
This paper was prepared to inform the OECD Competition Committee's first dedicated discussion of agentic AI and its implications for competition policy. It builds on several strands of its earlier work, as well as broader work across the OECD:
Analysis of algorithms and algorithmic competition considered how automated pricing and decision systems may generate efficiencies while also affecting transparency, personalised conduct and collusion between firms, including collusion instigated by machines rather than humans (OECD, 2017[7]; 2023[8]; 2025[9]).
Work on AI markets and competition has examined access to data, models, computing resources and infrastructure, as well as vertical linkages and potential barriers to entry (OECD, 2024[10]; 2025[11]).
Related work has considered how AI adoption may affect costs, entry, innovation and competition in downstream markets, and how digitalisation can connect consumer agency with competitive outcomes (OECD, 2026[12]; 2025[2]).
Elsewhere in the OECD, work has examined the agentic AI landscape and its conceptual foundations (OECD, 2026[1]), as well as the use of agentic AI in organisations (OECD, 2026[13]), amongst others.
This paper also draws on recent competition authority analysis, including the French Competition Authority’s examination of competitive conditions in the sector (see Box 2) and the United Kingdom Competition and Markets Authority’s work on the implications of agentic AI for consumers (Autorité de la concurrence, 2026[14]; CMA, 2026[15]). Bringing these strands together, it considers how greater delegation, decisions linked across multiple steps and interactions between AI agents may change the operation of familiar competition mechanisms or raise additional questions. Given the early stage of development, the aim is to identify relevant mechanisms, conditions and policy considerations rather than predict a particular market structure or set of infringements.
Section 2 explains the principal characteristics of agentic AI systems and reviews emerging market developments. Section 3 examines competition in their supply, including barriers to entry and expansion and potential sources of concentration. Section 4 considers how their use may affect competition in other markets, including through consumer delegation, changing intermediation, exclusion and co-ordination. Section 5 draws out the implications for policymakers and competition authorities.