The Brain / Capabilities / Stakes routing
REASONSmall things run fast.
Consequential things stop for you.
Reversible work runs fast and cheap; irreversible, client-facing or financial work gets the strongest reasoning and stops in front of a person.
Three actions, three treatments
Sorting overnight mail runs unasked
Classification is reversible and invisible outside the company; a wrong label gets changed and nothing happened.
A client reply is drafted, and stops
The draft is ready in seconds, and the gap between ready and sent is a person reading it.
A scope call gets the expensive reasoning
Whether a change is inside the agreement is costly to get wrong, so it is presented with the clause attached.
Three rules, none about business outcomes
Route on reversibility and blast radius
Can this be undone, and who sees it if wrong; an urgent-sounding message cannot buy its way past a confirm.
The stop is on the action
Anything that sends, deletes or shares is marked once, so the same send stops for a person from a chat, a briefing or a schedule.
The AI cannot approve its own action
Consent is minted where a person clicked, and an approval in the AI’s own output is not consent.
Three honest limits
Not an approval workflow; one human yes, no matrix.
It does not judge whether the action was wise.
The confirm cannot be switched off to go faster.
Updated 15 September 2026 · WE_AINA
Book a Diagnostic Sprint
Bring your list of what may run alone, and watch it run.