It earns delegation the way a person does.
A chief of staff you can't trust is just noise. So the foundation under everything else is a simple promise: nothing happens silently, everything is reversible, and the system only takes on more once it's proven its judgement.
Propose, then act
Every action the Chief of Staff wants to take arrives as a concrete proposal you approve, edit or reject — often in a single tap, in a batch. Nothing changes behind your back, and every change is written to a history you can always look back through and undo.
The approval loop
This is the substrate the whole system rides on. A meeting commitment, a re-ranking, a tidied tag — each is a typed proposal. You stay the decision-maker; the system does the legwork of spotting and drafting.
Graduated autonomy
Trust is set per type of action, not as one big switch. Low-stakes, reversible, internal things — tidying a tag, resurfacing a parked item — can graduate to “just do it, and tell me.” Anything irreversible, external, or high-stakes stays a proposal, always. And every autonomous action still writes to history, so it's reviewable and undoable.
The capabilities in this layer
Propose & approve
Every action a typed proposal; one-tap, batched; nothing invisible.
LiveGraduated autonomy
Per-action trust with a stakes classifier; reversible work can be handled automatically.
LiveAlignment scorecard
An honest mirror of how much effort is going to genuinely strategic work.
LiveAgentic execution
The endgame — the system doing the low-value busywork, not just managing it.
RoadmapA record that looks after itself.
Four layers — raw material in, judgement applied, surfaced where you can act, on a foundation you can trust. Together they take the admin out of staying on top of work, and leave the decisions where they belong.