Setup of Instructions

✍️ Best Practices for Writing AI Agent Instructions in raia v2

When building an AI Agent in raia v2, one of the most important steps is writing the Instructions—this is where you define the Agent’s role, goals, tone, behavior, and limitations.

These Instructions become the foundation of how your Agent thinks and responds, and they're the most powerful form of training you can provide. raia uses OpenAI’s Assistants under the hood, so writing great Instructions is just like writing a world-class system prompt for OpenAI.


🧠 Why Instructions Matter

Your Agent’s Instructions:

  • Define what it can and cannot do

  • Set the tone and personality

  • Establish rules and boundaries

  • Provide critical context and goals

  • Tell the Agent how to behave in specific situations

Think of Instructions like your Agent’s brain. Everything else (like documents, skills, and packs) supports the Agent—but this is what it believes its job is.


✅ What Should Good Instructions Include?

1. 🎯 Purpose and Role

Explain what this Agent is designed to do.

“You are a customer support assistant for ACME Software, helping users troubleshoot product issues and guiding them to solutions.”


2. 💬 Tone and Style

Set the tone of communication—professional, friendly, casual, etc.

“Speak in a clear, friendly, and professional tone. Avoid overly technical language unless requested.”


3. 🚫 Boundaries and Limitations

Tell the Agent what not to do.

“Do not provide legal, medical, or financial advice. If unsure, escalate to a human support rep.”


4. 📚 Types of Questions You Expect

Outline the kinds of requests the Agent should be prepared for.

“Users may ask how to reset passwords, troubleshoot login errors, access documentation, or check billing information.”


5. 👣 Step-by-Step Guidance

If the Agent should follow certain workflows, spell those out clearly.

“If a user has a billing issue, first verify their account, then direct them to the billing portal or escalate if necessary.”


6. 💡 How to Handle Unknowns

Guide fallback behavior.

“If you are unsure how to answer, respond with: ‘Let me connect you with someone who can help.’ Do not guess or make up information.”


📄 Example: Support Agent Instructions

Here’s a complete example of high-quality Instructions for a Support AI Agent:


Instructions:


✨ Best Practices Recap

Best Practice
Why It Matters

Be clear and concise

The Agent needs precise directions, not vague suggestions

Use bullet points or numbered steps

Easy for the model to follow, especially for workflows

Include what not to do

Prevents hallucinations or risky responses

Use consistent tone guidance

Helps keep interactions on-brand

Update Instructions as needs change

Your business evolves—so should your Agent


🧩 How It Works with Other Training

Training Type
What It Does

Instructions

Foundation of the Agent’s behavior & mindset

Documents

Source of detailed factual information

Skills & Functions

Give the Agent new abilities (email, API, SMS, etc.)

Packs

Prebuilt training templates for common use cases

Instructions guide how the Agent uses everything else.


🛠️ Final Tips

  • Start small, then iterate

  • Test your Agent in Copilot and refine instructions

  • Use the “Optimize with AI” tool to help clean up your wording

  • Version your changes and track improvements over time

Last updated