Spotlight on agentic AI | From procurement to governance

AI systems are beginning to do more than generate content. Across enterprises, they are starting to plan, decide and act within live business workflows - initiating tasks, adapting strategies and, in some cases, coordinating with other agents or external systems. These are “agentic AI” systems.

McKinsey’s State of AI survey¹ reports that 23% of respondents say their organisations are scaling an agentic AI system in at least one business function. 

That shift from generative AI adoption to agentic AI is already affecting how organisations procure, deploy and govern AI. It raises legal and governance questions that generative AI frameworks were not designed to address. It also creates risks that are harder to govern precisely because the systems involved are harder to observe and control.

Compared with traditional generative AI, agentic deployment changes the nature of the legal and governance questions. For example:

• Access: what data, tools and systems can it reach?
• Memory: what does it retain between sessions or across interactions?
• Authority: what is it allowed to do?
• Discretion: how much freedom does it have to decide the next step?
• Evidence: what can you prove afterwards about what it accessed and what it did?

Articles and Resources:

This series sets out the key concepts, legal considerations and governance questions that arise when organisations deploy agentic AI - from initial procurement through to ongoing oversight.

• Understanding Agentic AI - what it is, how it works, and how it differs from generative AI.
• From Content Risk to Conduct Risk: Identifying Agentic AI Risks - the main legal and compliance risks arising when systems can act, not just generate.
• Contracting for Agentic AI - the contractual issues to address when procuring and deploying agentic solutions.
• Governance and Oversight - how governance, approvals and monitoring need to evolve when systems can initiate actions.

Given the pace of change in AI technology, these articles reflect the position as at the date of publication.