I was in the room the first time I watched an organization discover it had no answer to the simplest question a board can ask after something breaks: who owned this.
The system in question was not an AI agent. This was years before that term meant anything outside a research lab. It was an automated decisioning tool that had quietly approved a batch of transactions it should have flagged, and by the time anyone noticed, the cost had already cleared six figures. The post-incident meeting filled a conference room with the people you would expect: the technology lead who had built the tool, the vendor relationship manager who had licensed the underlying model, the business unit head whose team relied on its output, and the risk officer whose job was theoretically to have caught this before it happened.
I asked the question directly. Who owned this system. Four people started answering at once, and every answer was partially true and entirely insufficient. The technology lead owned the build. The vendor manager owned the contract. The business unit owned the outcome. Risk owned the framework that was supposed to have caught the failure and had not. Four true answers add up to zero accountable people, because accountability is not divisible. The moment four people can each say “that part was mine,” the organization has designed a structure where no single person has to answer for the whole.
I have watched this exact pattern for thirty years, in rooms that had nothing to do with software. A board increases oversight after a performance concern, adding committees and reporting requirements, and each addition feels like progress. What actually happened is that the board filled an accountability gap with control, because control is easier to build than the conversation control is meant to replace. Nobody sat down with the one person who should have owned the outcome and stated, plainly, what needed to be done, what authority they had to do it, what success looked like, and what would happen if it failed. I call this the Accountability Contract Model, and in every version of this failure I have watched, that conversation is the thing that never happened. Committees get built. Frameworks get written. The contract does not get made, because the contract requires naming one person, and naming one person means that person can be wrong in public.
That is the part organizations avoid. Naming five owners costs nothing, because no single reputation is on the line. Naming one owner is uncomfortable, because it puts a name where a diagram used to be. Boards and executive teams will build extraordinary amounts of governance architecture to avoid that discomfort, and every version of it produces the same result: a system that can fail without anyone specific failing.
The agentic systems entering enterprises now do not give an organization the luxury of a conference room and four hours to sort out who owned what. They act, and the action is already taken before the four-person debate could begin. The rooms I sat in years ago had time to argue about ownership after the fact. The systems being deployed today do not leave that window. The old failure just got faster.
Activity is not accountability, and most organizations still cannot tell the difference. A dashboard with five stakeholders attached to it looks like governance. It is not governance. It is the appearance of coverage, purchased at the cost of anyone being able to answer the one question that matters when something goes wrong. The organizations that will handle this new wave of autonomous systems are not the ones with the most committees. They are the ones willing to write one name on the whiteboard before the system goes live, not after it fails.
The next generation of officers in these organizations will inherit one of two things. Either a name that was written down before anyone needed it, something the people who come after can point to and build on, or an argument about who should have been in the room, repeated at every incident review from now on. The leaders who write the name down first are not responding to a regulator or a failure that already happened. They are building it from conviction rather than crisis, and their successors will never know the difference between the two, because the name was already on the board before anyone needed it.
This is developed in the Agentic AI Governance Playbook, where the Single Owner Protocol lays out exactly what one name on that whiteboard is required to know, control, and be able to prove.
Glenn E. Daniels II, Touch Stone Publishers