At the 13th Competition Day in Hamburg, organized by Solve Economics, Jan Krämer took part in a high-level panel on digital market power. Together with Sabrina Frank (Idealo), Carolin Marx (Hogan Lovells), and Niklas Brüggemann (Latham & Watkins), and moderated by Mitja Kleczka (Solve Economics), he discussed how fair competition in digital markets can best be ensured—through traditional ex post enforcement of competition law or through ex ante regulation such as the Digital Markets Act (DMA).
In his intervention, Prof. Krämer focused in particular on AI chatbots and AI agents, and on the question of how—and whether—these systems should be addressed under the DMA going forward.
Agentic AI: the next wave
Prof. Krämer outlined the next wave of so-called agentic AI systems, characterized by (i) significantly greater autonomy, (ii) the ability to adapt and learn from experience—including access to new data and data-driven network effects, (iii) perception of their environment (e.g., via sensors, actuators, and “tools”), and (iv) the capacity to plan and “reason.” These developments could amount to a paradigm shift, such as “appless phones” or the emergence of new super-gatekeepers, and are far less science fiction than often assumed. Given the rapid pace of change, regulation could very quickly arrive too late.
Competition concerns in two phases
He distinguished two main phases of potential competition problems:
- Foreclosure of an AI agent (early phase):
Incumbent gatekeepers may prevent the emergence of strong new AI players through input foreclosure (restricted access to data, foundation models, compute, or on-device sensors) and distribution foreclosure (pre-installation, defaults, self-preferencing, and bundling with proprietary services). Deep integration with operating systems is likely to be essential for many agent functionalities. - Foreclosure by an AI agent (later phase):
Once powerful agents are established, familiar super-gatekeeper concerns arise, including self-preferencing, refusal to deal, and strong user lock-in effects.
Can the DMA address these risks?
Prof. Krämer argued that the DMA is surprisingly future-proof in this respect. In the first phase, the DMA can effectively address foreclosure risks through interoperability obligations, choice screens, de-installation rights, and data-related requirements—provided that the relevant actors are appropriately designated.
For the second phase, he suggested treating AI agents or “virtual assistants” consistently as a relevant category in their own right. While current developments also exhibit features of browsers and search engines, targeted adjustments to the DMA framework—for example regarding the definition of business users and data-sharing obligations analogous to “click and query data”—could mitigate key competition concerns. Crucially, regulation should also prevent forced integration with a specific AI agent, similar to the DMA’s approach to alternative browser engines.
Conclusion
Prof. Krämer concluded that agentic AI will be among the most important technologies of the future, likely intensifying some well-known competition problems in digital markets. Although the DMA was originally conceived with existing markets in mind, it nevertheless proves remarkably adaptable—not least due to the (admittedly late) inclusion of the virtual assistant category. Given the speed of technological development, there are strong arguments for early regulatory engagement, as history shows that intervention often comes only after market structures have become entrenched and difficult to reverse.