Software to establish, keep and act from decisions.
When your company decides something, where does the deciding happen?
Companies are starting to let AI agents do work that people did. That work waits on questions the company settles every day: is this allowed, is this ready, is this right.
The deciding has lived in people. The software got the result: approved, released, done. When agents do the work, the deciding can no longer live only in the people involved, even where a person still supplies the judgment.
The deciding happened in a person.
Before the machines, a computer was a person who did the calculating. Your company still has a computer for deciding, and it is still a person.
That person knows what the company needs before a question counts as answered: which facts, from which sources, which rules apply, whose judgment is needed. They know whether that was met this time, what showed it, and when the answer would stop holding: a policy changes, a fact turns out to be wrong.
People are also slow, few and answerable. They hesitate on the big one, ask a colleague or send it up. They can be asked why, and they answer for what follows. Nobody designed that as a control. It came with the person.
The software has what the person left behind.
Ask what the software has, and the answers come fast. Someone approved it. It went through the process. It’s in the record. They were allowed to. None of them is wrong. Where the cost of being wrong was high, the company asked for more: the plan, the certified claim, the checklist signed line by line. The name, the completed steps, the event, the permission and the signed sheet are what the software has. The deciding is not in any one of them, or in all of them together.
The company is handing the deciding to agents.
An agent can read the policy, gather the evidence, apply the rules and propose an answer. It may know more about the question than the person did. And there will be more: more work worth doing, more agents supplying judgment, and more systems acting on answers they took no part in reaching.
What the person carried was never a thing of its own, so it does not come with the job. Nobody expects a calculator to hesitate. When the calculating left the human computers, it had to become something a machine could use. The deciding has to now.
An answer is not enough.
Intelligence can answer the question. The company decides what it takes to rely on the answer.
A model can produce a convincing assessment while something the company requires is still missing. It can also be right for the wrong reason: it misreads the policy, and a fact it never looked at makes the answer correct anyway. The company has to be able to tell an answer apart from something it can rely on, and whoever relies on it next, person or system, needs to know what was established and what established it.
Judgment comes from a person or a model. A rule gives a result. What it takes to rely on any of them belongs to the company. A more capable model gets trusted with more consequential work. Companies did not remove signing authorities because the people holding them were capable.
The deciding has to become a thing.
What the person carried for one question, asked once, is a decision. It is not the moment someone chose. It is four things kept together: the question, its answer, what the company required before that answer could count, and what met it this time. It is one thing, apart from the person and apart from the work that was waiting on the answer.
The company sets what it requires for that kind of question, not the person who answers it this time. Sometimes it is already written down, as a policy or a checklist. The rest happened in the person, and of that the software got the yes.
A decision has to be established, kept and acted from.
The decision exists because what the company required was met, not because someone clicked. Until then there is an open question that says what is still missing. Meeting what the company requires does not prove the answer right. It establishes what the company can stand behind, and what a later person or system can check. That is the first thing: established.
The decision has to still be there afterward, with the answer connected to what the company required and what met it. One piece of work rests on several decisions, and one decision is what the next twenty rely on. Today the next piece of work asks again, or copies the answer without what was under it. A decision that exists apart from the work can be found, asked whether it still holds and relied on where it does, and when something under it changes, the company can find what depended on it. That is the second thing: kept.
An act has to rest on the decisions that support it. Permission says who may act. It cannot by itself establish that this act should happen. The system that acts can check that support at the moment it acts. Another person or system can check it later, elsewhere, without having taken part in the deciding. That is the third thing: acted from.
Ordinant is where the decision happens.
Ordinant is software in which a decision happens. The decision is established when what your company requires for that question is met, and every piece of work that rests on it can find it there, whoever is doing the work.
People and agents supply judgment. Rules supply results. The decision is the company’s.
Pick one piece of work an agent could do for your company. What it is waiting for is not a better agent. It is the decisions it depends on, as things. Bring us that piece of work.