# How ZazzyAgent AI Agents Work: Knowledge, Prompts & Actions A ZazzyAgent AI Agent is more than a chatbot that generates text. An AI Agent can understand what a customer is asking, use your business knowledge, collect information, decide when an action is required, perform supported actions, and hand a conversation to another agent or human when needed. The easiest way to understand an AI Agent is through four parts: **System Prompt + Knowledge + Actions + Agent Configuration** Each part has a different job. ## 1\. The System Prompt The System Prompt is the agent's main instruction set. It tells the agent: * Who it is * What its job is * How it should communicate * What information it should collect * What it should not do * When it should use knowledge * When it should use an action * When it should transfer the customer * How it should behave when information is missing The current underlying AI Agent documentation describes the System Prompt as the main control layer for role, behavior, instructions and actions. Think of it as the operating instructions you would give to a new employee. ## 2\. Knowledge The knowledge source gives the agent information it can use when answering customers. Your knowledge can cover: * Products * Services * Pricing information * FAQs * Shipping * Returns * Policies * Company information * Documentation * Website content * Business processes The underlying platform supports knowledge campaigns containing sources such as FAQs, URLs and files, with other supported sources depending on configuration. ### Knowledge and instructions are different This distinction is very important. **Knowledge tells the agent what information is available.** **The System Prompt tells the agent what to do with that information.** Example: Your knowledge says: > Standard delivery takes 3–5 business days. Your System Prompt might say: > When a customer asks about delivery times, answer using the shipping information in the knowledge source. Do not invent delivery estimates. Both are needed. ## 3\. Actions Actions allow the AI Agent to do something rather than only send a reply. Depending on your ZazzyAgent configuration, an agent can be given actions such as: * Add a label * Remove a label * Assign a sequence * Save a custom field * Call an HTTP API * Use Shopify * Use WooCommerce * Trigger a bot flow * Assign a human team member * Assign a team role * Transfer to another AI Agent * Mark the conversation as solved * Block genuine spam * Skip a reply * Create an internal note * Set a follow-up reminder The current underlying AI Agent architecture supports these kinds of operational actions. ## 4\. Agent Configuration Creating an AI Agent is not always enough by itself. The agent also needs to be enabled and configured for the conversation-routing setup. The current underlying platform separates agent creation from Agent Configuration, where active agents, routing and related settings are controlled. ## Example: Ecommerce Sales Agent Imagine an online clothing store. A customer says: > I need a black shirt for a wedding. What do you recommend? The AI Agent can: 1. Understand that this is a product-recommendation request. 2. Search the connected product knowledge or store. 3. Recommend relevant products. 4. Answer follow-up questions about size or style. 5. Save useful lead information. 6. Apply a lead label. 7. Continue the conversation. If the customer then says: > I already placed an order last week. Where is it? The agent may transfer the conversation to a support/order-status agent rather than continuing as a sales agent. That is where specialized AI agents become useful. ## One agent or multiple agents? You don't always need multiple agents. A single agent can be enough when the business has a simple use case. For example: > FAQ + basic lead capture A multi-agent setup becomes useful when different types of conversations require different knowledge or actions. You could have: **Sales Agent** Handles products, recommendations and purchasing. **Support Agent** Handles orders, shipping and refunds. **Lead Qualification Agent** Collects business requirements and qualifies prospects. **Appointment Agent** Handles appointment-related conversations. The current AI Agent architecture supports specialized agents and routing between them. ## What happens when an action runs? The AI model does not directly modify your business system by itself. A simplified process is: **Customer message** ↓ **AI understands the request** ↓ **AI decides whether an action is needed** ↓ **ZazzyAgent executes the configured action** ↓ **Result is returned when applicable** ↓ **AI explains the result to the customer** The underlying platform documents this separation between AI decision-making and actual platform execution. ## Why actions need clear instructions Giving an agent an action does not automatically tell it when the action should happen. This is why the System Prompt matters. Bad instruction: > Use Shopify. Better: > When a customer asks about an existing order, use the Shopify order lookup action. Do not guess the order status. The agent needs a clear relationship between: **Condition → Action → Customer response** That pattern should be used throughout your System Prompt. ## A good AI Agent is focused Avoid creating one enormous agent that tries to handle every possible task. A focused agent is easier to train, easier to test and easier to troubleshoot. A good Sales Agent might handle: * Product discovery * Product recommendations * Pricing questions * Buying questions * Lead qualification A Support Agent might handle: * Order status * Shipping * Returns * Refunds * Complaints Keeping responsibilities clear makes the agent's behavior easier to control. ## The basic AI Agent formula A reliable ZazzyAgent AI Agent usually looks like this: **Role** Who am I? **Goal** What am I trying to accomplish? **Knowledge** What information can I use? **Conversation rules** How should I talk and ask questions? **Actions** What can I do? **Restrictions** What must I never invent, promise or change? **Handoff** When should another agent or a human take over? That becomes the foundation for your AI Agent system prompt. zazzyagent