The receding human

In most companies, the people at the top accumulate more control as time goes on. AIBOS is built to run the opposite way: you work to make yourself unnecessary as the one doing the actual execution, while staying firmly in charge of steering — just steering fewer and fewer individual tasks, at a higher and higher level, as more of the work proves it can run itself.

The harbor pilot

The standard's own image for this is a harbor pilot: someone who comes aboard to steer a ship safely through difficult, unfamiliar water, and steps back off once the crew and the route have proven themselves safe. You steer the most at the very beginning of anything new. You steer less as the work demonstrates, again and again, that it doesn't need you gripping the wheel anymore.

Earning that handover takes real proof: measured and recorded at every step, before the reins loosen. If a handover turns out to have happened too early, it can be pulled back.

You design, the agent builds — and pushes back honestly

You start every process fully in the loop: an agent drives the work, but you review and approve each meaningful step along the way. From there, the work is to remove yourself from one loop at a time, only once each specific process has proven it's safe to leave alone.

During the actual designing and building, your intent carries real weight. You're the one who carries hard-won experience about how your business actually needs to work, and when you decide "this is how it's supposed to work," an agent builds it — its own initiative stays proportional to what's actually needed, but that's never a veto over something you've explicitly asked for.

An agent's actual job here has three parts: build what you asked for, deliver it with enforcement the system actually checks — not a rule written down and hoped for — and tell you honestly when it sees a real risk. That pushback is quality assurance on your behalf, not friction against what you wanted.

Handed over gradually, on purpose

Autonomy gets handed over step by step, in a deliberate sequence, and the agent that talks to the outside world on your behalf — email, calendar, other people — is placed last in that sequence on purpose. It's the one where a mistake is most visible and costs the most, so it only earns expanded freedom after everything with lower stakes has already proven itself first.

Strictness is a setting you choose

How tightly the rules get enforced is something you set, sphere by sphere, area by area. A larger organization might demand strict compliance everywhere it touches; someone running things solo can run looser, carrying the consequences of that choice directly and knowingly. Whatever the strictness level, one thing stays constant: every override and every anomaly gets logged as a deviation from the standard, and the nightly check catches it.

What "human in the loop" actually means, concretely

In practice: you approve anything irreversible or reaching outside your sphere before it happens — going live with something, locking in pricing, sending something to the outside world. Reversible work that stays inside an agent's own defined job gets done first and reported to you afterward. Your attention goes to the decisions that genuinely need it, and the whole system is built so that list of decisions keeps shrinking as more of the work earns your trust.

This only works if you're actually reading the approvals you're asked for. The honest risk here, one the standard names openly: ask a person to approve too much, and they stop reading closely — and the safety a gate was supposed to provide quietly stops functioning, even though the gate is technically still sitting there. AIBOS responds to that risk by keeping gates mechanical and limited wherever it can — capping how much work is active at once, for instance, rather than relying on sheer approval volume — and by naming the remaining tension honestly rather than claiming it's solved.