How to make sure you get the best results from atoms agents
✅ Good Examples | ❌ Bad Examples |
---|---|
Collect Customer Name | Node 1 |
Verify Payment Status | Step 3 |
Send Order Confirmation | Miscellaneous Task |
Retrieve Order Details | Process A |
Check User Subscription | Task 42 |
❌ Bad Example (Vague) | ✅ Good Example (Explicit) |
---|---|
“Ask the user what they want. If they mention an order, ask for details. If they ask about a product, give information." | "Introduce yourself as a customer support agent. Be professional, warm, and helpful. Inform the customer that you can assist with product inquiries, order tracking, and support requests. Begin by asking how you can help today." |
"End the conversation politely." | "Thank the customer for considering the service. Inform them that they are welcome to call back anytime to ask queries. Close the conversation professionally." |
"Confirm the booking details." | "Summarize the details you have collected: the date, time, number of guests, and any special requests. Ask the customer to confirm if everything is correct or if they need to make any changes. Note down any changes.” |
❌ Bad Practice | ✅ Good Practice |
---|---|
AI generates a random response for a refund policy. | Use Static Text to display the exact refund policy message. |
AI freely rephrases legal disclaimers. | Set disclaimers as Static Text to maintain accuracy. |
AI dynamically creates a privacy notice. | Use Static Text for legal and compliance-related messages. |
❌ Bad Practice | ✅ Good Practice |
---|---|
Webhooks are integrated without testing. | Use API testing tools (Postman, cURL) before deployment. |
API responses are not checked for formatting. | Ensure API responses are correctly formatted. |
Missing or incorrect variable extraction. | Verify that all necessary variables are passed and extracted correctly. |
No error handling in case of API failure. | Implement error handling to avoid workflow failures. |
❌ Bad Practice | ✅ Good Practice |
---|---|
Extracts the entire API response without filtering. | Only extract necessary fields (e.g., order_id , status ). |
Uses vague field names (data1 , valueX ). | Use clear, meaningful field names (e.g., customer_name , subscription_status ). |
Doesn’t validate extracted data. | Test the workflow to verify extracted values are correct. |