Working with AI Agents
Communicating Effectively with Agents
The way you communicate with an AI agent dramatically affects the quality of results you get. Here's how to get the most from your interactions.
The Anatomy of a Good Request
1. Be Specific About the Goal
Vague: "Help me with the sales report" Specific: "Create a summary of Q3 sales by region, highlighting any territory with >15% growth compared to Q2"
The specific request gives the agent clear success criteria.
2. Provide Relevant Context
Without context: "Write a follow-up email" With context: "Write a follow-up email to a prospect I met at the AI conference last week. They were interested in our enterprise plan but concerned about implementation time."
Context helps the agent make appropriate choices.
3. Specify Format and Constraints
Open-ended: "Give me information about competitors" Constrained: "Create a one-page comparison table of our top 3 competitors, focusing on pricing, key features, and target market. Use bullet points, max 5 items per cell."
Constraints prevent the agent from going in unhelpful directions.
4. Indicate Quality Standards
Basic: "Draft a proposal" Quality-focused: "Draft a proposal using formal business language, suitable for a Fortune 500 CFO audience. Include specific data points where possible, and flag any assumptions."
This helps the agent calibrate its approach.
Common Patterns That Work
The Role Assignment
"Act as a financial analyst reviewing this investment proposal. What questions would you ask before recommending approval?"
The Step-by-Step
"First, research X. Then, analyze the findings for Y. Finally, present recommendations in format Z."
The Example-Based
"Here's an example of what I'm looking for: [example]. Now create something similar for [new context]."
The Iterative Refinement
"Let's start with a rough draft. I'll give feedback, then we'll refine."
What to Avoid
Being Too Vague
"Make this better" doesn't give the agent direction. Better: "Make this more concise" or "Add more technical detail."
Overloading Requests
Too many requirements at once can dilute quality. Break complex requests into stages.
Assuming Knowledge
Don't assume the agent knows your company's internal terminology, processes, or recent events. Provide context.
Expecting Perfection First Try
Agents work best with iteration. Plan for a review-and-refine cycle rather than expecting final output immediately.
Quick Tips
- Start simple, add complexity — Begin with core request, add constraints as needed
- Use concrete examples — Show what you want rather than just describing it
- Check assumptions early — Ask the agent to confirm its understanding before proceeding
- Give feedback — "This is close, but..." helps the agent adjust
:::