We’re entering a new era of artificial intelligence (“AI”)
While many are already using generative AI ("GenAI") to write content, produce code, create visuals, or generate ideas, the next step is agentic AI - systems that can plan and carry out complex tasks on their own. Think AI that can autonomously build and manage software, or designs game environments personalised for each player. This shift opens up huge possibilities but also introduces new legal challenges. As AI becomes more autonomous, existing frameworks struggle to address its decision-making capabilities. In this report, we examine the legal implications for the European Union when AI acts as a genuinely self-directed agent and its impact across different sectors.
What is agentic AI?
Autonomy in problem solving
Agentic AI refers to AI systems that are given agency - the ability to autonomously perceive, reason, act, and learn in pursuit of a goal with minimal human input. In other words, an agentic AI system can make decisions and take actions on its own to achieve complex objectives, adapting its behaviour based on the context.
These systems combine the generative and understanding capabilities of advanced AI, such as LLM, with decision-making logic and tool use. So, they’re not just generating content, they’re creating it themselves.
Key characteristics of agentic AI include:
- Autonomy in decision-making/proactiveness: They can assess situations and determine next steps with little to no human input. For example, once you tell it, “organise my calendar and book travel for a meeting”, it can carry out the necessary sub-tasks on its own.
- Goal-driven problem solving: Agentic agents use a sense-plan-act cycle - observe data, reason through it, execute actions, and learn from results. This loop helps them tackle complex tasks and adapt as things evolve.
Difference between generative AI and agentic AI
Traditional GenAI models, such as early chatbots and image generators, usually operate in a single-step input/output format without continuing goals. GenAI is largely reactive, focusing on creating content (text, images, etc.) in response to prompts.
Agentic AI, by contrast, uses AI agents that operate within a goal-oriented feedback loop. Rather than simply answering a question or producing a one-off output, these systems manage and execute entire processes.
For example, while a GenAI model might write an email when prompted, an agentic AI could go several steps further - drafting the email, sending it, scheduling the meeting based on the reply, and updating your calendar - all with minimal human input. This shift makes agentic AI feel more like an intelligent assistant or operator, rather than just an analytical tool.
Explore sector-specific use cases and legal challenges
To show the real-world potential and complexity of agentic AI, we’ve outlined four use cases across different sectors.
Each case demonstrates how these systems move beyond simple generative content production by adding layers of autonomy and execution that significantly increase both utility and legal complexity. While the underlying content generation generally presents familiar legal issues like those of traditional GenAI, the added layers of autonomy and execution introduce a whole new set of challenges.
We highlight key legal challenges for each selected use case, though this is not an exhaustive list.
