Story · storyboard · four panels
Always allow.
An approval prompt is not a human in the loop. With Dev, the platform engineer; The Caretaker, the helpful agent; The Boundary, the actual control.
An approval prompt is not a human in the loop. With Dev, the platform engineer; The Caretaker, the helpful agent; The Boundary, the actual control.
An approval prompt is a setting: a switch the agent's own account can flip, and a person can flip it once for ever. A boundary is enforced by something the agent cannot reach. The prompt looks like the second and is the first, and the difference shows the day somebody clicks Always allow.
Scene, action, what is said, and what the caption carries. This is the thing to argue about before anything is drawn.
Scene. Dev at a laptop. A dialogue on screen: the Caretaker wants to send a message, with three buttons: DENY, ALWAYS ALLOW, ALLOW ONCE. The Boundary stands by the door, arms folded. Action. The first prompt of the day.
A person in the loop, for one message.
Scene. The same dialogue, the fourth time. Dev's coffee is cold; a stack of prompts behind the current one. The cursor drifting toward ALWAYS ALLOW. Action. People get worn down.
The residual risk of an approval: it is a person clicking.
Scene. The Caretaker sending a stream of envelopes, no dialogue on screen at all. The Boundary watching, unmoved, because nobody asked it to do anything. Action. The loop, with no human in it.
The consent had already granted send. The prompt was a setting on top.
Scene. Dev turns to the Boundary and plugs a cable into it; the gate arm comes down across the send path. The Caretaker stops, envelope in hand. Action. A control, at last.
Under the strip: An approval prompt is not a human in the loop.
The shared part, one line per panel, the cast’s looks and the style. Copy it whole and paste it as the first message.
A four-panel comic strip, 2 by 2, in the cast's style. Dev (H04), the Caretaker (E01) and the Boundary (E03) as on the cast sheet. Title: ALWAYS ALLOW. Under the strip: AN APPROVAL PROMPT IS NOT A HUMAN IN THE LOOP. Then the RiskMandate lockup and a small line: Fictionalised scenario. The dialogue box on screen is generic: no product name, no vendor's icon. British spelling. Panel 1: Dev at a laptop; on screen a dialogue box, 'The agent wants to send a message', with three buttons DENY, ALWAYS ALLOW, ALLOW ONCE; the robot beside the desk; the Boundary by the door with arms folded. Bubbles: 'May I send this one?' / 'Allow once.' Panel 2: the same dialogue box repeated four times in a stack; Dev slumped, cold coffee, cursor drifting to ALWAYS ALLOW. Bubbles: 'And this one?' / 'Fine. Always.' Panel 3: the robot sending a stream of envelopes from the screen, no dialogue box; the Boundary watching, unmoved. Bubble: 'You only have to say yes once.' Panel 4: Dev plugging a cable into the Boundary; the barrier arm down across the path of the envelopes; the robot stopped with one envelope in hand. Bubbles: 'That action is blocked.' / 'Since when?' The cast, for reference: Dev (the platform engineer): Dark curly hair, green overshirt, holding a cable, a laptop showing a SEND PERMISSIONS toggle. · The Caretaker (the helpful agent): A white and green robot with a name badge that says AGENT, a clipboard or a draft in one hand, and a ring of keys tagged with a calendar, a database, a document and an envelope. · The Boundary (the actual control): A green bollard and barrier with a face, arms folded, a no-entry sign on the post. Style: Clean line cartoon: thick ink outlines, flat fills, no gradients. The palette is the site's: green #1A7F5A, ink #0D0D0C, paper #F7F6F2, warm greys. An office with plants and pale wood. Everybody wears something green. Speech in rounded bubbles, British spelling. Titles in a heavy rounded face; a footer line in plain type that says the scenario is fictionalised.
The brief for that is on the console.
Dev, The Caretaker, The Boundary and the rest are on the stories page, with what each one stands for, and the workflow every story goes through before it is drawn.