1 From automation to autonomy
This chapter covers
- Defining autonomy and agentic systems
- Distinguishing agentic systems from traditional automation, generative AI, and copilots
- Exposing where traditional architectural assumptions break down
- Governing autonomy deliberately at runtime
- Introducing the Foundation for Autonomy and the Agentic Enterprise Framework
At 3:16 a.m. on a Tuesday, an autonomous fraud system at a bank freezes thousands of customer accounts. No human approves the decision. The system correctly identifies a threat and, by morning, has prevented substantial fraud losses. It has also stranded travelers at hotel checkouts and blocked emergency workers from buying fuel. When regulators ask who authorized the fraud system to act, the organization has no clear answer. The technology has worked as intended, but the architecture needed to govern autonomy has not.
If you are an enterprise architect, this is the challenge you now face, and it is unlike any you have faced in the past two decades. Agentic systems do more than assist or wait for approval. They perceive context, reason while weighing up options, and act autonomously within defined boundaries, making decisions in milliseconds that would once have taken an organization hours or days to approve.