AVAILABLE January 2027
When an AI agent can make decisions and act,
policy becomes part of how the work runs. This book gives business and technology leaders a practical method for deciding where agents can help, agreeing how they should operate, and building those decisions into everyday workflows.
Agents act, so policy must run inside the work.
When an AI agent can make decisions and act, the organization’s policies need to shape how that work happens, including the actions the agent can take and the approvals it needs.
The decisions are harder than they look.
Giving a customer-service agent permission to change an appointment still leaves open which customers it can move, whose time it can commit, and when it needs a person’s approval. Those decisions touch several parts of a business and need an agreed owner.
The work doesn’t end at launch.
Once an agent is running, someone must review the results and decide whether it should keep operating on the same terms.
The book includes four tools that help teams identify unresolved decisions, document agreements and organize the work of putting policy into operation.
Readiness assessment
Decide whether a proposed agent is worth deploying.
Policy decision register
Record what was agreed and who owns each decision.
Three-day policy sprint
Bring the teams involved together to work through the open decisions.
First-ninety-days plan
Organize the work after launch.
From Chapter 2: The Agent Policy Gap
Recently I worked with a global pharmaceutical company that deployed an AI agent to speed up content creation for primary healthcare providers (HCPs) around one of its therapies. The marketing team used the AI agent to summarize clinical studies, identify content opportunities, and draft educational materials. Work that once took hours was now produced in minutes. The team named the agent “Pia,” short for Pharmaceutical Information Agent. The name stuck because it read like a person, a product, and a new colleague all at once.
The pilot was designed to test whether the agent could make an existing information task faster and soon saw that it did. Then, the organization widened access and put the tool in the hands of HCPs through one of its agent-supported portals. However, widening the access also widened the questions. A system that had been evaluated on its ability to retrieve and organize approved information was now being asked questions that required it to interpret that information. The shift was incremental enough that it did not look like a new use case until the agent answered a critical question.
An HCP asked how the therapy might perform under a particular set of circumstances for a select patient population, and the agent gave a sound answer based on the information available to it. Nothing about the exchange raised a flag, and nobody gave it a second thought. Months later, during a routine compliance review, someone raised a question about who made the decision that the agent was allowed to answer that question. The room went quiet. Marketing assumed Medical Affairs had set the boundaries. Medical Affairs assumed Legal had reviewed the use case. Legal assumed Technology had implemented the necessary controls. Technology assumed the business had defined the requirements. Everyone believed someone else had made the decision, and nobody could say who.
I wish I could say this was unusual. When everyone in a cross-functional room confidently believes somebody else owns a decision, it is highly likely that nobody does. The names, technologies, and org charts change, but the pattern stays the same. Responsibility is assumed rather than assigned, and the cause is rarely negligence, bad intent, or a lack of expertise. The cause is the distance between how fast a capability spreads and how slowly the organization’s agreements about it catch up.
In December 2025, I sat in a workshop at a multinational beverage company and watched six senior leaders spend twenty minutes debating who owned an emerging technology initiative. Eventually I asked, “If we needed to shut it down this afternoon, who could make that decision?” The all too familiar silence enveloped the room, and that silence was more useful than the whole preceding discussion. A policy gap usually reveals itself through a question nobody can answer.
Excerpt from Chapter 2, The Power of Agentic AI Policy: A Practical Guide to Autonomous Systems Your Organization Can Trust, by Kristina Podnar. Forthcoming January 19, 2027. © 2026 NativeTrust Consulting, LLC. All rights reserved.
For leaders putting AI agents to work, and the colleagues who work alongside them.
The focus is on the operating decisions your organization needs to make.
“A practical and effective roadmap for leaders at every stage of AI maturity.”
“Not simply what AI agents can do, but what they should be allowed to do, under whose authority, and with what accountability.”
“I recommend the book to every executive navigating the shift to agentic AI.”
“AI agents can acquire authority through permissions, thresholds and vendor settings before an organisation has agreed who owns the decision.”
“An outstanding and actionable guide for leaders.”
“A must-have practical guide for our agentic times.”
“Chapters 6 through 9 are critical and indispensable for any leader looking to build a resilient, trustworthy, and competitive organization in the age of agentic AI.”
“A grounded perspective on the organisational choices and controls that agentic systems require.”
January, 2027, in hardcover, paperback and digital editions.
Yes. Journalists, reviewers and interview hosts can request a digital review copy through the Get Involved form at the top of this page. Review PDFs will be available at the end of November 2026.
Sign up through the Get Involved form at the top of this page and we’ll email you release and purchasing information.
Yes. For copies for a leadership team, governance committee or client group, choose bulk orders in the Get Involved form at the top of this page.
%202017-10-13%2022-55-13.png){kind=logo}
