When AI Starts Acting
What changes when systems move from tools to agents
As AI systems begin to initiate actions, trigger workflows and operate across tools with increasing autonomy, organisations are encountering a new operational reality. What was previously a decision support tool is increasingly beginning to act within operational workflows across products, services and internal systems.
This shift is not arriving as a single technological moment. It is emerging gradually through integrations, orchestration layers and product capabilities that allow systems to take action within live environments. In many organisations this behaviour is already present, often unevenly and sometimes without a clear internal language for what is happening.
As systems move closer to execution rather than recommendation, new questions begin to surface. Where should delegation sit between human and machine. What does meaningful oversight look like when systems are able to act across tools. And how should organisations think about accountability as systems begin to take action within operational workflows.
This roundtable will examine how organisations are approaching these questions today. The discussion will explore where agentic behaviour is already appearing in practice, how delegation and intervention are being considered as autonomy increases, and what early signals organisations should be paying attention to now.
The aim of the session is to develop a clearer shared view of how agentic behaviour is emerging in practice and what this means for organisations now.
Insights from the discussion will also help shape the next phase of work led by the UKAI Agentic AI Working Group, as the group explores practical approaches to oversight, accountability and operational control in agentic systems.
Discussion themes:
-
When AI moves from automation to agency
-
Human oversight and delegation in agentic systems
-
Oversight models for systems acting across workflows
-
Accountability when AI executes decisions
-
Early signals organisations should be tracking as autonomy increases
-
Intervention and control when agentic systems act unexpectedly
