When an agent acts, someone still has to answer for it
A year of pilots proved the models can do the work. The question that decides which agent programs survive is duller and harder — who is accountable when one gets it wrong.
The bottleneck on enterprise agents was never whether they could do the work. It is whether anyone can be held responsible when they do it badly.
The field is about to thin
That distinction is about to thin the field. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027 — not because the technology underdelivers, but because of escalating costs, unclear value, and inadequate risk controls. Only about one in five organizations has a mature governance model for the agents it is already running. The capability is racing ahead of the accountability, and the gap is where the projects die.
What governance actually means
It is worth being precise about what "governance" means here, because the word has been drained of meaning by overuse. It is not a policy document. It is the answer to a specific question: when an agent takes an action — moves money, changes a record, sends a message to a customer — whose name is on that decision, and can you reconstruct why it happened? If the answer is "the model decided," you do not have an agent. You have unowned risk wearing the costume of automation.
The market is not helping
Gartner reckons only about 130 of the thousands of vendors selling "agentic AI" are the real thing; the rest is agent washing — a renamed chatbot, an RPA script with a new logo. A CIO's first job in 2026 is subtraction: refusing the demos that dissolve on contact with an audit question.
The organizations pulling ahead are not deploying more AI. They are deploying it with the boring scaffolding in place from day one — identity, logging, code review, an incident path — so that every agent action has an owner and a trail. That scaffolding is unglamorous and it is the whole game.
Treat the accountable owner as a hard prerequisite, not a phase-two nicety. An agent no one can answer for is not an employee you hired. It is a liability you deployed.