CASE FILE Agent Loops §1/7

Operation

AGENT LOOPS

Classification EYES ONLY

Subject area SEQUENTIAL

Document date ██████████

Distribution RESTRICTED

Topic · The Question

A single prompt answers once and stops. How does an AI turn one goal into a chain of actions that keeps going until the job is actually done?

Make the loop work.

When you tackle a big task, what steps do you go through without thinking?
What does an AI do today when one answer isn't enough?
Have you watched an AI 'think' through steps? What did it look like?
How would a machine know when it's finished a job?
What's your guess: how does an AI decide what to do next?
When has a plan of yours needed changing halfway through?

▶ OPEN DOSSIER