The output is good enough to ship, and nobody in the room can say who is answering for it. Three people assume it was checked by one of the others.
AI responsibility is the question of who carries the consequence of an AI-assisted output. AI accountability is whether that person can be identified after the fact. A responsible AI framework is the document stating the principles both should follow. All three describe intent for an organisation, not behaviour for a task. This method works one level down: each recurring task gets a mode — automate, assist, supervise, practice or refuse — and a named human owner for every mode that is not full automation. Not a compliance product, not a certification, not legal advice.
Voice, judgment and responsibility drift together: the work stops sounding like the people doing it, nobody can reconstruct why a choice was made, and no name is attached before the output leaves.
Take one task and write seven lines: what AI may prepare, what it may propose, what it must not decide, what the human verifies, the evidence required, the accountable owner, and one verdict — delegate, prepare, verify, refuse or escalate. Free, no account, no payment, nothing leaves the page.
Task: draft and send the weekly client update. May prepare: pull delivery numbers, assemble the timeline, draft a neutral summary. May propose: two framings for the slipped milestone. Must not decide: whether and how we tell the client the deadline moves. Human verifies: every figure and date against source, plus any sentence containing a commitment. Evidence: source file and date in the draft, reviewer initials in the send checklist. Owner: the account lead, by name. Verdict: prepare.
The reading comes from What Remains Human?, Release 02 of the World Under Pressure audio series.
The paid field kit is The Human Layer Kit — 79 €: ten modules, printable core cards, worksheets and a one-page personal operating protocol. Not legal or compliance advice, and not a certification.
Other situations: answering a decision under pressure · seeing the infrastructure behind the event.