Do not use agents just because the word sounds advanced. Use an agent when the task is multi-step, has observable progress, can use tools, and has objective success criteria.
- Is the task multi-step?
- Can intermediate decisions be made safely?
- Are tools available and scoped?
- Can success be objectively verified?
- What happens if the agent makes a mistake?
Agent fit prompt
Decide whether this should be handled as chat, a workflow, or an agent. Explain the reason, risks, needed tools, checkpoints and verification criteria.
Task:
[paste]