Judging levels of AI autonomy

You use a suitable autonomy level to surface alternatives, exceptions and reader questions beyond your initial plan.

Using the assistant to produce work beyond what you could have produced alone, rather than the same work faster.

You brief the agent with its goal, reader, source facts, constraints and required output before judging autonomy.

Giving the assistant the reader, purpose and source facts the work depends on, and the form, length and contents you need back.

You turn autonomy evidence into a usable recommendation, explaining the level, limits and human approval needed.

Turning the assistant's output into something a colleague could use: editing it, combining it and adding original thinking.

You check the agent's claims about capability, reliability and failure against supplied evidence before accepting its autonomy recommendation.

Checking the assistant's claims, figures and assumptions against the source before the work is used.

You keep personal, restricted and dangerous details out of the agent's brief and output while setting safe operating boundaries.

Handling confidential and personal data, seeing compliance, ethical and reputational risk, and declining unsafe requests.

You decide what the agent may do, what stays with a person, and who approves each consequential decision.

Deciding what to hand to the assistant and what to keep, and at what stakes, before any output exists.