AI agents (or agentic AI) represent the next step in the evolution of artificial intelligence. Unlike a chatbot that answers a single query in isolation, an AI agent can break a complex task down into sub-steps, choose the right tools and sources, execute sequential actions and adjust its strategy based on intermediate results. It marks a shift from "question and answer" to autonomous problem solving.
In the legal field, AI agents open up considerable possibilities. Jus Mundi launched Jus AI 2 in September 2025, an agent specialised in international law that measured a 125% improvement in relevance compared with a conventional LLM. A legal agent can, for example, take in a complex question, identify the applicable texts, search for relevant case law, cross-reference scholarly sources, and generate a structured summary — all in an autonomous, iterative way.
This autonomy nonetheless raises crucial questions of liability. Who is responsible for an action taken by an AI agent without an intermediate human review? How do you guarantee the traceability of decisions? Professional ethics frameworks and the AI Act will have to adapt to this new reality, where AI no longer simply answers but acts proactively.