# How to Create an AI Agent in ZazzyAgent ZazzyAgent AI Agents are designed to handle customer conversations with a defined responsibility. Instead of creating one AI that tries to do everything, you can create focused agents such as: * Sales Agent * Customer Support Agent * Lead Qualification Agent * Order Status Agent * Appointment Agent This guide explains how to create your first one. ## Before creating an AI Agent Prepare three things: ### 1\. Decide what the agent is responsible for Keep the responsibility narrow. Good: > Help customers choose the right products and answer pre-purchase questions. Less useful: > Handle everything related to the company. ### 2\. Prepare your business knowledge Collect the information the agent will need. For a sales agent this might include: * Product descriptions * Pricing * Product specifications * Size information * Delivery information * Return policy * Frequently asked questions ### 3\. Decide which actions the agent needs Only give the agent actions it actually needs. A product sales agent may need: * Product search * Lead label * Custom-field saving A support agent may need: * Order lookup * Human assignment * Internal notes * Solving conversations ## Create the AI Agent The current underlying AI Agent interface uses an AI Agents area with a **Create Agent** option, followed by agent configuration fields. In ZazzyAgent, open the AI Agent section and choose the option to create a new agent. You will configure the agent's basic information. ## Give the agent a clear name Use a name that describes its responsibility. Good examples: * Sales Agent * Support Agent * Order Status Agent * Lead Qualification Agent * Appointment Agent Avoid vague names such as: * AI 1 * Bot * Assistant * New Agent A clear name becomes even more useful when you eventually create multiple agents. ## Add a description The description should explain what the agent handles. Example: > Handles product questions, recommendations and pre-purchase sales conversations. Another: > Handles order status, shipping, returns and customer support requests. Keep the description focused. ## Write the System Prompt The System Prompt is where you define the agent's behavior. At minimum, tell it: * Who it is * What it is responsible for * Who it serves * How it should communicate * What information it should collect * How it should use knowledge * When it should use actions * What it must not do * When it should transfer to a human or another agent The current AI Agent documentation uses the System Prompt as the central configuration for the agent's role, behavior, data collection, knowledge use and actions. ## Connect the right knowledge source Select the knowledge source that contains the information this agent needs. Don't connect unrelated information simply because it is available. For example: **Sales Agent knowledge** * Products * Sizes * Pricing * Shipping * Returns * Promotions **Support Agent knowledge** * Order policy * Shipping policy * Returns * Refunds * Support FAQs The knowledge source provides information; the System Prompt tells the agent how that information should be used. ## Add only the actions the agent needs Suppose you are creating a Sales Agent. It might need: > Save lead information > Add interested-lead label > Recommend products It does not necessarily need: > Cancel orders Giving an agent unnecessary capabilities makes your automation harder to control. ## Save the agent Save the configuration. At this point, the agent has been created. That does not necessarily mean it is already handling customer conversations. ## Enable the agent The current underlying platform has a separate Agent Configuration area where agents are enabled and routing is configured. Open your agent configuration and make sure the agent is active for the intended channel/bot. If you are using multiple agents, configure how conversations should reach the correct agent. ## Test the agent Do not immediately send traffic to it without testing. Try questions from the real customer perspective. For a Sales Agent: > Do you have black shirts? > What size should I buy? > How much is this? > Can I return it? > I want something under ₹2,000. Then test unexpected questions: > Where is my previous order? > I want to talk to a human. > Hello > Thanks > ??? The agent should respond appropriately in every case. ## A simple first-agent setup For your first test, keep it simple. ### Agent name Sales Agent ### Responsibility Help customers choose products and answer pre-purchase questions. ### Knowledge Product information + shipping + returns. ### Actions Product recommendation + lead label + custom-field saving. ### Handoff Transfer order issues to the support agent. This gives you a manageable first AI Agent that you can improve before building a much more advanced setup. ## When should you create another AI Agent? Create another agent when the task requires a different responsibility, knowledge source or workflow. For example: **Sales Agent → Support Agent** The customer starts by asking about a product. Later they ask: > Where is my order? That is no longer a sales question. A specialized setup can transfer the conversation to an Order or Support Agent. This type of specialized multi-agent workflow is supported by the underlying AI Agent system. ## Final checklist Before enabling your agent: * \[ \] Name is clear * \[ \] Responsibility is specific * \[ \] System Prompt is complete * \[ \] Correct knowledge source is connected * \[ \] Only required actions are available * \[ \] Human handoff is defined * \[ \] Agent is enabled * \[ \] Test conversations are complete [zazzyagent.com](http://zazzyagent.com)