AI Agents in Action Foundations for Evaluation and Governance 2025

Introduction AI agents are gradually becoming embedded in an increasing number of tasks, workflows and use cases that span cloud and edge computing, leading the way to more widespread adoption. As the transition from prototyping to deployment accelerates, current adoption remains concentrated among early adopters. According to a recent global survey of executives, 82% of organizations plan to integrate agents within the next one to three years, indicating that most efforts are still in the planning or pilot phase,1 while moving towards wider adoption. The concept of software agents has been studied for decades in fields such as robotics, autonomous systems and distributed computing. What is different today is the rise of data-driven models, particularly generative artificial intelligence (AI) and large language models (LLMs), which are enabling the emergence of a new generation of LLM-based agents. These systems can generate plans, simulate reasoning and adapt their behaviour through feedback mechanisms in ways that were previously not possible. This evolution has sparked a new wave of experimentation, with researchers and companies rapidly creating prototypes of agents in various fields. This report focuses mainly on LLM-based agents (“AI agents” is sometimes used in short), whose growing capabilities create both significant opportunities for adoption and a new set of challenges in governance and safety.AI agents are shifting from prototypes to deployment, bringing both transformative opportunities and novel governance challenges. Foundations for the responsible adoption of AI agents FIGURE 1 Functional classification2 Define the agent’s roleEvaluation and governance3 Scale with confidenceTechnical foundations1 Lay the groundwork AI Agents in Action: Foundations for Evaluation and Governance 5