Writing Effective Prompts for AI Agent Creation

When creating an AI Agent using SysAid’s AI Agent Builder, providing a well-structured and detailed prompt ensures the agent is generated with the right capabilities, parameters, and behaviors.

This guide will help you craft an effective prompt with minimal iteration when defining your AI Agent.

Key Components of an AI Agent Prompt

Your prompt should include the following sections to ensure clarity and functionality:

You can view the examples below to understand each section better.

📌 Best Practices for Writing Prompts

When creating your own AI Agent, follow these best practices to ensure the best results:

✅ Be Clear and Specific – Clearly describe the agent’s purpose and expected behavior
✅ Define Parameters – Specify the inputs the agent should consider
✅ List Relevant Data Sources – Identify where the agent should pull information from
✅ Describe the Expected Actions – Explain what the agent should do with the data
✅ Handle Errors Gracefully – Ensure the agent provides meaningful error messages
✅ Provide Sample Questions – Show examples of the queries the agent should understand

By following this structure, you can create powerful AI Agents tailored to your IT operations.

Tip:

You can paste the example prompt directly into the AI Agent Builder chatbot and ask it to modify the prompt to fit your needs ✨

Example: "Analyze Duplicate Service Requests" AI Agent

Here’s a structured prompt for an AI Agent that identifies duplicate service records in an IT service desk.

Example Prompt:

Purpose: The agent should analyze open service records to identify potential duplicates based on user, asset, and content similarity. It should provide recommendations for merging service records with detailed analysis and confidence scores.

Limitations:
- It should focus on open/active service records by default, but I want the option to include closed service records if needed.
- It should be able to analyze anywhere from 10 to 200 service records at a time to find duplicates. If no range was specified, analyze a maximum of 100 service records.
- It should look back up to 168 hours (7 days) to check for similar service records. If no time range was specified, look back 7 days.
- I want to set a similarity threshold (between 0.5 and 0.95) to control how strict the duplicate matching should be.

Parameters:
The agent should be able to compare service records based on:
- User (e.g., same requester submitted multiple service records)
- Asset (e.g., multiple service records about the same laptop or printer)
- Primary and Secondary Categories (e.g., service records about the same type of issue)
- Assigned Group (e.g., duplicate issues assigned to different teams)
- Location, Priority, Impact, and Urgency
- Before running the analysis, it should group service records based on the user, asset, category, location, or assigned group—whichever is most relevant.

Data Sources:
It should use prebuilt SysAid analytics data and SysAid’s LLM model to analyze service record similarities.

Actions:
The agent should perform the following actions:
1. Retrieve service records based on the filters I set.
2. Group related service records before running a similarity check.
3. Use LLM to analyze service records descriptions and identify duplicates.
4. Provide a detailed report with:
  - A similarity score between service records.
  - Common themes in duplicate service records.
  - A recommendation on whether they should be merged.
  - A clear rationale explaining why service records were flagged as duplicates.

Error handling: 
The agent should handle errors gracefully and provide detailed error messages if any issues occur during execution.

Sample Questions: 
- Show me potential duplicate service records in the current queue.
- Find service records submitted by the same user for the same asset.
- Which service records have similar descriptions and should be merged?
- Analyze open service records for duplicate issues.
- List service records that might be reporting the same problem.

Example: "Find Assets with Missing Critical Information" AI Agent

Here’s a structured prompt for an AI Agent that finds Assets with Missing Critical Information.

Example Prompt:

Purpose:
The AI Agent should help the IT team maintain complete and accurate asset records by identifying assets with missing critical information. It should generate a response listing assets with incomplete details and notify the relevant IT personnel.

Limitations:
- The agent should only check active assets—decommissioned or archived assets should be excluded.
- It should focus on specific missing fields that impact operations, such as serial numbers, warranty details, assigned users, or locations.
- The agent should not modify asset records—only identify missing information and generate reports.

Parameters:
The AI Agent should identify assets missing the following critical fields:
- Serial Number
- Warranty Expiration Date
- Assigned User or Department
- Location
- Purchase Date
- Asset Category (e.g., Laptop, Server, Network Device)

Additional customization options:
- Filter by Asset Type (e.g., only show missing details for network devices or end-user laptops).
- Filter by Department (e.g., only check assets assigned to a specific team).

Data Sources:
- SysAid Asset Management Database
- Hardware & Software Inventory Records
- Warranty & Purchase Information

Actions:
1. Scan asset records to identify missing critical information.
2. Generate a structured report listing all assets with missing details.
3. If a missing detail was recently updated, exclude it from future reports unless still incomplete.
4. Allow users to manually request an up-to-date report on demand.

Error Handling:
- If an asset record is missing from the database, notify the IT team instead of including it in the report.
- If a data source is temporarily unavailable, retry before alerting the IT team.
- If no assets are missing information, send a confirmation report stating "All asset records are complete.


Sample Questions the AI Agent Should Handle:
- Which assets are missing serial numbers?
- Generate a list of laptops without an assigned user.
- Show me all network devices missing warranty expiration dates.
- What are the most common missing fields in our asset database?
- Send me a weekly report of assets with incomplete details.
- Can I see today’s report instead of waiting for the weekly one?