The single biggest factor in agent quality is the specificity of your request. Vague prompts get vague results.
A simple structure that works
- State the goal in one sentence.
- Define the scope — which records, which time window, which segments.
- Specify the output format — table, draft email, summary, action.
- Add constraints — what to exclude, what to confirm before acting.
Example: weak vs strong prompt
“Send a follow-up to leads who haven't replied.”
— Weak — too vague
“Draft individual follow-ups to all B2B leads in our distributor segment who opened our last campaign, did not reply, and are not in an active opportunity. Use a friendly tone, reference the SKU they showed interest in, and keep each message under 80 words. Show me the drafts before sending.”
— Strong — actionable
The strong version produces predictable, reviewable output. The weak version produces something the agent guesses at — and a workflow you cannot trust.