Prompt Library: Understanding the Task

Every prompt from the course, grouped by what you're trying to do. Swap the brackets for your own details.

Mapping Your Agent's Loop

Help me think through how an agent would handle a task, step by step. THE TASK: [what it

should accomplish]. SYSTEMS IT COULD REACH: [data and actions available]. WHAT A HUMAN DOES

TODAY: [walk through it]. Write a plausible step-by-step trace for a typical case, what it

thinks, what tool it calls, what it observes, why it continues. Then write a trace for an

awkward case where the first thing it tries doesn't work. How many steps does each take, and

where does the count vary most? Name three points where it could plausibly choose wrongly and

say what it would do instead. And what should the give-up condition be, with what handed back

to the human?

Mapping Your Connections

Help me plan the connections an agent would need. WHAT IT SHOULD DO: [description]. SYSTEMS WE

HAVE: [list]. For each system: does it need to READ, WRITE, or both? If it writes, is that

reversible and what's the worst outcome of a wrong write? What specific capability does it

actually need, be narrow, not 'access to the CRM' but the two or three specific things. Which

could plausibly have an off-the-shelf connection already, and which would need building? Then

rank the whole list: which connection delivers most of the value, and could we ship something

useful with only the top two or three?