Autonomy, with a human always in command

The most common worry about autonomous systems is also the most reasonable one: if a machine decides for itself, who is responsible? Our answer is simple, and we build around it. A person is always in command.

Machines execute; people decide

An autonomous system can carry out a task without someone steering it moment to moment. That's the point — it's what makes autonomy useful. But carrying out a task and deciding whether it should happen are two different things. The second belongs to people, and our systems are designed to keep it there.

Guardrails, not guesswork

In practice that means clear limits on what a system will and won't do, a person able to supervise and intervene at any time, and behaviour that stays inside defined bounds. Autonomy operates on a leash of our choosing — long or short as the task requires.

Trustworthy autonomy is autonomy you can predict, supervise and switch off.

Assessment closes the loop

Responsibility doesn't end when a run does. We keep records and assess every run honestly — what went right, what went wrong, and why — then feed that back into the design. Accountability you can inspect after the fact is what makes autonomy something an institution can actually adopt.

This isn't a constraint we tolerate; it's a principle we design toward. It's how autonomy earns its place.