Traditional software waits to be asked. An agent decides to act. That single difference changes everything about how compliance applies.
Most systems in production today follow a simple pattern: a human triggers an action, the system responds, the interaction ends. The request and response happen in a single moment. Nothing persists. Nothing continues without input.
Agents break this model entirely. An agent perceives its environment, reasons about what to do next, acts across one or more external systems, and observes the results — often without a human involved at any step. The components are consistent across implementations: a model that reasons, a memory that persists state across steps, tools that interact with external systems, and a feedback loop that continues until the task is complete.