Creating Agents
This page walks you through creating and configuring AI agents in Simple Agent Designer.
Accessing the Agent List
- From the Agent Role Center, choose AI Agents under the Agent Designer section, or
- Search for AI Agents in the Tell Me search bar
The Worker List page opens, showing all configured agents.
Creating a New Agent
- On the Worker List page, choose New to create a blank row
- Fill in the following fields:
Required Fields
| Field | Description | Example |
|---|---|---|
| Agent ID | A unique code identifying this agent (up to 20 characters) | ANALYZER |
| Type | The worker type that determines the agent's capabilities | AI |
| Title | A descriptive name for the agent's role | Invoice Analyzer |
| AI Provider | Which AI service powers this agent | ChatGPT OpenAI |
| LLM Model | The specific AI model to use | gpt-5-mini |
Optional Fields
| Field | Description | Default |
|---|---|---|
| Use Tools | Enable tool/function calling for this agent | No |
| General Agent | Mark as the default agent for the interactive chat | No |
Configuring Agent Settings
Choosing a Worker Type
The worker type determines what the agent can do. Select the type that matches the agent's intended role:
| Type | Caption | Use When |
|---|---|---|
| AI | AI | General-purpose tasks: analysis, summarization, routing |
| DataAI | Data AI | Looking up and querying Business Central data |
| EntryAI | Journal Entry AI | Creating general ledger journal entries |
| SalesEntryAI | Sales Entry AI | Creating sales documents (quotes, orders, invoices) |
| PurchaseEntryAI | Purchase Entry AI | Creating purchase documents (quotes, orders, invoices) |
| ALAI | AL Runner AI | Executing AL code snippets |
| DataModifier | Data Modifier AI | Updating existing Business Central records |
For most workflows, start with the AI type for your primary agent. Add specialized types when you need agents that create documents or modify data.
Choosing an AI Provider and Model
- Set AI Provider to one of your configured providers (ChatGPT OpenAI, Anthropic, Azure OpenAI, or LM Studio)
- Set LLM Model to the specific model name
If the AI Provider has not been configured with a valid endpoint and API key, the agent will fail when processing workloads. See Provider Setup for configuration instructions.
Enabling Tools
Set Use Tools to Yes if the agent needs to:
- Query Business Central tables
- Create or update records
- Run reports
- Create journal lines or documents
When tools are enabled, the agent's AI requests include tool definitions, and the AI can request tool execution as part of its response. See Tool System for details.
Setting the General Agent
One agent should be marked as the General Agent (displayed as "General Agent" in the UI). This agent is used by the interactive chat on the Role Center. Only one agent should have this flag enabled.
Writing Agent Instructions
Each agent has custom instructions stored in a blob field. To edit them:
- Select the agent in the Worker List
- Choose Edit Instructions from the action bar
- The Instruction Editor opens
- Write your instructions in the editor
- Close the editor to save
What to Include in Instructions
Good agent instructions typically include:
- Role definition — "You are a financial analyst specializing in accounts payable"
- Task description — "Your job is to analyze incoming invoices and extract key information"
- Output format — "Return your analysis as a JSON object with fields: vendor, amount, date, category"
- Rules and constraints — "Never create journal entries with amounts exceeding 10,000"
- Context about your business — "We use cost centers 100-199 for operations and 200-299 for sales"
Example Instructions
Invoice Analyzer Agent:
You are an accounts payable specialist. When given an invoice or receipt:
1. Extract the vendor name, invoice number, date, and total amount
2. Identify the appropriate expense category from our chart of accounts
3. If the amount exceeds $5,000, flag it for management review
4. Return a structured summary with your findings
Our main expense categories are:
- 6100: Office Supplies
- 6200: Travel and Entertainment
- 6300: Professional Services
- 6400: Software and Subscriptions
Data Lookup Agent:
You help other agents find information in Business Central.
When asked about a customer, look up their balance, credit limit, and payment terms.
When asked about a vendor, look up their outstanding invoices and payment status.
Always return data in a structured format that other agents can parse.
For detailed guidance, see Agent Instructions.
Testing Your Agent
After creating an agent, test it:
Quick Test via Interactive Chat
- Mark the agent as General Agent (or temporarily set it)
- Go to the Role Center
- Use the interactive chat to send a test message
- Verify the agent responds correctly
Test via Workload
- Create a test workload with a simple task description
- Assign it to a workload type that uses your new agent
- Process the workload
- Review results in the Process Chat or Activity History
Managing Agents
Editing an Existing Agent
Simply update the fields directly in the Worker List, or select the agent and choose Edit Instructions to modify its custom instructions.
Deleting an Agent
- Select the agent in the Worker List
- Press Delete or choose the delete action
- Confirm the deletion
Note: Deleting an agent that is referenced as the Initial Worker in a Workload Type will cause that type to fail when processing. Update the Workload Type first.
Viewing Other Agents' Instructions
Agents can see summaries of other agents in the system. The GetAIs() function provides a formatted list of all other agents and their instructions, which is used internally when agents need to decide which peer to delegate to.
Tips and Best Practices
- Use descriptive Agent IDs that reflect the agent's role (e.g.,
INVOICEAI,DATAQUERY,GLPOSTER) - Keep agent instructions focused — one agent should do one thing well
- Start with tools disabled and enable them only when the agent needs to interact with BC data
- Test agents individually before combining them in multi-agent workflows
- Use cheaper models for simple tasks and more capable models for complex analysis
- Document what each agent does in its Title field for team reference