Before the mandate comes the will

Carlos Ocampo · Field Notes D31 · September 2026

An authorization can be perfectly recorded and still leave a question unanswered: under what conditions was that decision made?

Did the person know their options? Did they understand the consequences? Could they stop? Could they consult someone? Could they choose a different way out?

Human autonomy needs those conditions before it becomes a permission within a system.

In building an autonomous company, defining what an agent can execute also requires understanding how people decide what to delegate. An instruction can be well formulated and serve a purpose nobody stopped to question.

Before asking what the system can do, we need to ask what we want to do and why.

At D31, we propose making that reflection part of the design of the relationship between humans and agents. The 51/49 rule begins there: in preserving the human ability to form an intention, examine it and change it.

In our formulation of Human Centered Agent™, the quality of collaboration must also be examined through what it leaves with the person: the ability to understand, decide and act again. Completing a task can be useful; learning to formulate it and review the result also has value.

Consider an everyday decision. A director asks an agent to recommend how to reduce costs. The system presents a convincing option. Before authorizing it, the director needs to understand the assumptions behind it, which alternatives were left out and what consequences it would have for the people involved.

The director also needs room to reconsider the original question. Perhaps cost reduction is necessary. Perhaps the problem requires a review of priorities, coordination or revenue. The possibility of reframing the problem is part of that person's autonomy.

Through Self Human Centered Design™, we propose observing three conditions: that the intention remains recognizable to the person, that there is room to deliberate, and that they can regain control without becoming trapped in the relationship with the tool.

This requires concrete design and operational decisions. Making alternatives visible. Allowing questions. Identifying uncertainties. Enabling correction and exit. Preserving the possibility of turning to other people.

The speed of a response must leave room to understand it. The convenience of delegation must preserve the possibility of reviewing what was delegated.

We can design situations where refusing, correcting and leaving are possible, understandable and effective. These conditions allow us to observe whether the human sovereignty we declare has a practical expression.

The human 51% begins with the ability to want and decide before delegating.

When your team accepts an AI recommendation, what conditions does it have for questioning it and choosing another path?

© 2026 D31. All rights reserved.