Formulate effective AI prompts

Communicating effectively with AI agents is crucial for obtaining accurate and helpful responses. This guide will help you formulate clear, concise messages that maximize the agent's ability to assist you. These rules apply to runbooks as well.

Key principles for communication with AI

Large language models (LLMs) excel at interpreting natural language messages. However, to achieve the best results, follow a few basic principles of AI prompt engineering. These guidelines will help you avoid having to repeat requests due to parts of your query being ignored or misinterpreted.

Be specific and clear

Provide detailed, unambiguous instructions to the agent. Clearly state your objectives and expectations.

Instead of asking "Can you help me with a report?", try "I need a 500-word summary of Q3 sales performance, focusing on top-selling products and regions."

Use simple, concise language

Keep your messages brief and use straightforward language. This helps the agent process your request more effectively.

Specify the desired output format

Tell the agent how you want the information presented. This ensures you receive the results in the most useful format for your needs.

Example: "Please provide the results in a bulleted list with three main points and two sub-points for each."

Provide context

Give relevant background information to help the agent understand the broader context of your request.

Example: "I'm preparing for a meeting with our marketing team. We need to discuss our social media strategy for the next quarter."

Use a professional, direct tone

While politeness is appreciated, prioritize clarity and brevity in your communication. AI agents are tools designed to assist you efficiently.

Advanced tips for complex tasks

Break down complex requests

For complicated tasks, consider breaking them down into smaller, manageable steps. This allows you to guide the agent through your thought process.

Example:

1. "First, analyze the monthly sales data for the past year."
2. "Now, identify the top three performing products."
3. "Finally, create a table with the sales trend for these products."

Iterate and refine

Don't hesitate to follow up on the agent's responses. If the initial output isn't quite what you need, provide feedback and ask for adjustments.

Example: "That's a good start, but can you focus more on the impact of our new marketing campaign on these sales figures?"

Leverage the agent's strengths

Remember that AI agents excel at processing and analyzing large amounts of data quickly. Use this to your advantage by asking for comprehensive analyses or comparisons.

Example: "Compare our product features with those of our top three competitors and highlight our unique selling points."

Create productive runbooks

When you write a runbook for an AI agent, the principles are the same as for prompt engineering. Here are some additional tips for improved results:

• Be explicit: Stylistics don't matter as much as being explicit. Don't worry about repeating the sentence subject and object without replacing the subject and object of the sentence with a pronoun. It helps LLM understand.
• Repeat the rules: Your AI agent may often ignore a rule or two in your runbook. You can usually fix this by repeating the rule phrased differently in a different place in the runbook.
• Use Steps and Rules sections at the beginning. Only then add more details on the rules you outline in these sections.
• Put a brief outline of the rules and required steps at the beginning, right after the general Context and Objective sections.
• Use semantic formatting: Structure the runbook into clearly named sections, use bold for essential parts of sentences, and use ordered and unordered lists for enumerations.