# How to Enable and Configure an AI Agent in ZazzyAgent Creating an AI Agent is only the first part of setting up AI automation. You also need to tell [ZazzyAgent](http://zazzyagent.com): * Whether AI should be active. * Which AI Agent should handle conversations. * Whether AI should answer every message or only unmatched messages. * How the AI should operate within your customer conversations. These settings are managed separately from the Agent's System Prompt, Knowledge and Actions. ## AI Agent vs AI Configuration Think of it this way: **AI Agent** Defines what the agent is and how it should behave. **AI Configuration** Controls when that agent is active and how conversations are routed to it. For example: ```text AI Agent ↓ Role Knowledge System Prompt Actions ↓ AI Configuration ↓ Activation Response Mode Routing ``` ## Before you start You should already have an AI Agent. For example: > Instagram Lead Agent or: > Customer Support Agent See [How to Create an AI Agent in ZazzyAgent](https://blog.zazzyagent.com/how-to-create-an-ai-agent-in-zazzyagent). ## Step 1: Open AI Configuration Open the AI configuration area in ZazzyAgent. This is where you control the active AI behavior for your connected bot/channel. ## Step 2: Enable AI Agent Turn on the AI Agent setting. This allows AI functionality to operate for the relevant configuration. Creating an Agent without enabling the AI environment does not automatically make that Agent respond to customers. ## Step 3: Select the active Agent If you have multiple AI Agents, select the ones you want to make available. For example: * Lead Agent * Sales Agent * Support Agent You can keep an unfinished or experimental Agent inactive until it is ready. ## Step 4: Choose the AI response mode ZazzyAgent provides two important ways for AI to handle customer messages. ### AI Agent for All Queries In this mode, the AI can handle customer questions broadly rather than waiting for a message to miss your predefined automation. This is useful when you want AI to be the main conversational layer. ### AI as a Fallback Only In this mode, your predefined automation remains the primary system. AI steps in when a customer's message does not match the existing automation. For example: ```text Customer Message ↓ Predefined flow matches? ↙ ↘ Yes No ↓ ↓ Flow AI Fallback ``` This mode is useful when your bot already has structured flows for important tasks. ## When should you use All Queries? All Queries can work well when customers naturally ask questions instead of following menus. For example: > What services do you offer? > How much does your enterprise package cost? > Do you support businesses outside India? The AI can understand the questions and respond using its Knowledge and instructions. ## When should you use Fallback Only? Fallback Only is useful when your business depends on structured automation. For example: **Book Appointment** → Booking Flow **Track Order** → Order Flow **Request Quote** → Lead Flow If a customer asks something unexpected, AI can handle the question without taking control of the main structured journeys. ## All Queries vs Fallback Only | Mode | AI's role | | --- | --- | | All Queries | Primary conversational responder | | Fallback Only | Responds when predefined automation does not match | Neither mode is universally correct. Choose based on how your customer conversations work. ## Example: Structured support bot Suppose your chatbot has: **Track Order** **Refund** **Booking** **Contact Support** You may want: > Fallback Only This keeps the important processes deterministic. ## Example: General business assistant Suppose customers mostly ask open-ended questions. You may want: > All Queries This lets the AI act as the primary customer assistant. ## Multiple AI Agents If you've created specialized Agents, AI Configuration becomes more important. For example: ```text AI Configuration ↓ Active Agents ↙ ↓ ↘ Lead Sales Support ``` Routing can then determine which Agent should handle the conversation. ## Don't activate unfinished Agents Before activating an Agent, test: * System Prompt * Knowledge * Actions * Handoff behavior * Response quality An Agent that sounds good in one test conversation may still need work. ## Configure routing When several Agents are active, define which Agent should handle which type of conversation. For example: **New Lead** → Lead Agent **Pricing** → Sales Agent **Existing Customer** → Support Agent See [How to Build Multiple Instagram AI Agents in ZazzyAgent](https://blog.zazzyagent.com/multiple-instagram-ai-agents-zazzyagent). ## AI activation vs routing These are separate concepts. **AI enabled** means the AI system is active. **Agent selected** means that Agent is available. **Routing** determines which Agent handles a particular conversation. All three can matter in a multi-Agent setup. ## Test after changing the mode If you switch from: > Fallback Only to: > All Queries test your existing flows again. A customer message that previously entered a predefined flow may now be handled by AI depending on your configuration. ## Test unknown questions Send a message that does not correspond to an existing keyword or structured flow. This is especially important when using Fallback Only. You want to confirm that the AI fallback works as expected. ## Test known commands Also test your existing structured commands. For example: > booking > support > pricing Make sure your new AI configuration hasn't disrupted important workflows. ## Common problems ### Agent exists but doesn't reply Check whether AI is enabled. ### The wrong Agent responds Check active Agents and routing. ### AI responds when it shouldn't Check whether All Queries is enabled. ### AI doesn't handle unmatched messages Check Fallback Only and No Match configuration. ### A newly created Agent isn't being used Check whether it has been selected as an active Agent and whether routing points conversations to it. ## Recommended setup A practical structure for many businesses is: ```text Customer Message ↓ Specific structured task? ↙ ↘ Yes No ↓ ↓ Flow AI Agent ↓ Needs human? ↙ ↘ No Yes ↓ ↓ AI Human ``` This gives each system a clear responsibility. ## The key distinction The Agent defines: > **How should AI behave?** AI Configuration defines: > **When should that AI operate?** Keeping those two layers separate makes the entire AI system easier to manage.