Manj Chenna

The expertise layer · In plain terms

How an AI agent asks a human

Strip away the jargon and it is simple. At a decision it should not make alone, the agent stops, hands the question to a person with enough context to judge, waits for the answer, and only then acts. The simplicity is the point.

scroll to read
Manj Chenna · Founder, Sanctity · Building real human oversight of AI · Amsterdam

An AI agent asks a human by pausing at a hard decision, handing the question to a person along with the context they need, and waiting for the answer before it acts. That is the whole shape of it, described as an outcome rather than a mechanism. There are standard ways for software to make such a request, but the part that matters here is not the plumbing, it is the behaviour: the agent treats certain decisions as ones it must bring to a person, and it genuinely waits. That behaviour is what real human oversight of AI is for.

Agent reaches ahard callHands over thequestion and itscontextNamed persondecidesAgent acts on theanswerTHE PAUSE IS THE PRODUCT
Agent reaches a hard callHands over the question and its contextNamed person decidesAgent acts on the answerTHE PAUSE IS THE PRODUCT
The hard part is not the pause. It is handing over enough context that the person is deciding rather than guessing.

What a good ask includes

Enough for a person to actually judge: what is being decided, why it was flagged, and the information a reasonable human would need to decide well. An ask that dumps a yes-or-no with no context is how you manufacture rubber-stamping. A good ask respects the human's time and gives them what they need to use it.

Why this is more than a chatbot prompt

Because the agent is not chatting, it is deferring. It has identified a decision as one that belongs to a person, handed over real authority over the outcome, and bound itself to the answer. A chatbot asks you a question to continue its turn. An agent asking a human, in this sense, is pausing its own authority and placing the decision where the accountability belongs.

Read on

See when an AI agent should escalate and a portal for human judgment.