# For AI to do real work

We want AI and machines to do consequential work for institutions, and we want that work to be trusted. That trust is earned by how the work is constructed and what it leaves behind. These are the convictions Ordinant is built on.

## AI should do real work, not only prepare it for a person to finish.

Today AI mostly produces drafts, assessments and recommendations. A person remains the final step. That is how institutions decide the work is fit to proceed. Institutions should be able to let AI complete work that has decisions in it, and involve people where their standard requires a person.

## Trust should come from how the work was constructed, not from faith in whoever did it.

People, models and machines can all be wrong. An institution does not rely on consequential work merely because it trusts the person who did it. It relies on the evidence gathered, the checks passed and the record left behind. The same should hold for AI.

For consequential work, the institution should be able to know what was decided, what supported it, which requirements were met, what uncertainty remained, what was authorized and under whose authority, and what was actually done. So should those it answers to, including regulators, auditors, customers and counterparties.

## There must be room for judgment.

Where a rule can settle a question, use the rule. Much institutional work cannot be reduced to rules: whether evidence is relevant, whether an exception is justified, what to do with incomplete information. Judgment and uncertainty have to remain an accepted part of autonomous work, recorded rather than hidden, whether a person or a model supplies them.

## Decisions should belong to the institution and remain open to correction.

What the institution established, and why, should remain with it as people leave, models change and applications are replaced. It should remain open to examination and correction. Rigor does not make every decision right: evidence can prove wrong, judgments can be mistaken and requirements can change. The institution should be able to reconsider a decision and find the work that depended on it.

## Established decisions should remain usable in the work that depends on them.

Changing who does the work should not mean working out the institution's judgment again. When an established decision still applies, another person, agent, service or machine should be able to use it as it stands. Where it leads to action, the institution should retain what was authorized and what was actually done.

## The result should be more real work getting done.

As AI makes gathering evidence, contributing judgment and applying rigor cheaper, institutions can take on work that could never justify the human effort required. More rigorous deciding is not the goal. It is part of what lets work move from assistance to execution.

[Why Ordinant](/why-ordinant)

[What Ordinant does](/what-it-does)