Part 4 of 5: The Autonomy API—the three things an agent needs before you can trust it to decide

Autonomous work needs three things at once: the right to decide, a rule to decide by, and the facts the decision needs.

A decision can run without you only when three things are in place. Most organizations have never written down even one of them clearly, which is why their AI keeps asking permission.

Start with the symptom leaders feel but rarely name. In IBM’s 2026 survey of two thousand technology chiefs, 77 percent said AI adoption had already outrun their ability to govern it, and only 11 percent judged themselves fully ready for the coming scale of agent deployment. The fear is not that the tools are too weak. It is that they move faster than anyone’s ability to keep them inside the lines.

Both problems, the AI that underdelivers and the AI no one can govern, share a root and a fix. Any piece of work that cannot finish itself is short of one of exactly three things.

The right to decide. A rule to decide by. The facts the decision needs.

That is the whole list. Give a worker all three and you have handed them everything they need to act without being walked through your company every time. Call it the Autonomy API: the interface you hand to whoever does the work, a person today, an agent tomorrow.

Authority is the right to decide inside a stated boundary without asking anyone. Not the skill to decide, the sanctioned right. A person can know the correct call and still have to send it up the chain, because the call is not theirs to make.

Policy is a rule to decide by, stated sharply enough to draw a line. It does not mean a rule for every case. It means an intent clear enough that whoever meets a new situation can tell which side of the line it falls on.

Information is usable access to the specific facts a decision turns on. Not data sitting somewhere in the company. The fact, at hand, now. If answering the question means emailing ten people across four systems, the company has the information and the function does not.

A function needs all three at once. Two out of three does not get you two-thirds of the way there. It gets you a function that still cannot move, and an agent dropped into it stalls in exactly the place a person did, because you handed it the same gaps.

Now the part that decides whether any of this happens. Two of the three you can schedule. Policy is writing. Information is plumbing. The third, authority, is power, and power does not yield to a project plan. For one person to gain the right to decide, someone else has to stop being asked. That is why companies write excellent policies, build excellent dashboards, and still route every real decision through the same three desks.

So here is the test. Take your most frustrating function, the one where simple things take three weeks. Ask, in order: does the team have the right to decide, a clear rule to decide by, the facts at hand? Stop at the first no. That is your repair, and it is authority far more often than anyone wants to admit.

The model was never the missing piece. Three decisions a manager makes were: who may decide, against what rule, with what in front of them.

Next in the series: why an organization built this way holds the one advantage a competitor cannot buy from the same shelf.


Leave a Reply

Your email address will not be published. Required fields are marked *