Skip to content

Product systems · May 2026

Progressive autonomy: how software earns trust.

Trust is an onboarding problem. Even for software.

In short

Progressive autonomy means an automated system earns responsibility in stages, the way a new teammate does, instead of being trusted with everything at launch. The useful question is not “Can this system do it?” but “What is the smallest responsibility it can earn next?” There are four stages. Observe: the system shows what it notices and takes no action, so a person can check its view against reality. Practice: it proposes answers in a safe setting, where the strange cases and vague language surface. Assist: it does a bounded part of the work while a person stays close and can correct it, stop it or take over. Earn autonomy: only after it has been useful, consistent and easy to understand does it take on more, and always with a clear route back to a person. Making its limits visible is what makes the product credible.

We do not give a new teammate broad authority on their first day. We let them observe, practice in a safe place, make smaller decisions, and earn more responsibility over time. Yet we routinely ask people to do the opposite with automated systems: trust the output, hand over the action, hope for the best.

That is backwards. The useful question is not “Can this system do it?” It is “What is the smallest responsibility it can earn next?”

Four stages, not one leap.

1. Observe

Let the system show what it notices. No action yet. A person can compare its view with reality and decide whether the foundation is sound.

2. Practice

Let it propose an answer in a safe setting. This is where you find the strange cases, the vague language, and the moments that need more context.

3. Assist

Let it do a bounded part of the work while a person remains close. It should be easy to correct, stop, or take over.

4. Earn autonomy

Only after it has been useful, consistent, and easy to understand should it handle a larger responsibility. And even then, it needs a clear route back to a person.

The product lesson

This is not caution for caution’s sake. It is a better experience. People can learn what a system is good at without being asked to gamble on it. The product becomes more credible because it makes its limits visible.

Software earns trust the same way people do: small commitments, clear behavior, and an honest response when it reaches the edge of what it knows.