Managing AI Agents at Work

Your organization has AI agents. Can anyone say exactly what they can do, who owns the outcome, and how to stop them?

AI agents do more than generate answers. They can retrieve records, choose tools, send messages, update systems, and carry work across multiple steps. That promise also creates an operating problem: authority can spread faster than accountability.

Managing AI Agents at Work gives business, technology, risk, security, and operations leaders a practical method for controlling that authority without reducing governance to a policy binder or a one-time launch review.

At the center of the book is the BOUND Control Stack:
• Baseline the agent and register what is actually in use.
• Own the outcome through explicit accountability and decision rights.
• Understand the system's identities, permissions, tools, data, memory, and downstream reach.
• Name the tests, thresholds, evidence, and release decision.
• Design escalation, recovery, change control, and verified retirement.

Following one transparent composite case from discovery through retirement, Jamie R. Collins shows how to build an Agent Control File that remains useful after launch. You will learn how to:
• distinguish an assistant or fixed automation from action-taking software;
• write an Agent Task Contract with permitted and forbidden actions;
• expose the gap between intended scope and effective authority;
• match autonomy and human review to real-world consequence;
• evaluate full workflows, severe failures, recovery, and final system state;
• create production evidence without turning logging into a new privacy or security risk;
• define stop authority, preserve evidence, recover safely, and authorize restart; and
• reapprove material changes, revoke access, clean up dependencies, and prove retirement.

The included BOUND Control Kit provides reusable records, maps, decision forms, review agendas, and response templates. Framework orientation connects the method to established risk-management language while carefully avoiding false compliance claims.

This is not a book about predicting a distant autonomous future. It is an operating guide for the systems already entering customer service, finance, IT, procurement, HR, and other consequential workflows—and for the people who must remain able to explain, constrain, and control what those systems do.