# Understanding AI Agents for Customers 1.0 > By leveraging advanced artificial intelligence, AI Agents seamlessly resolve issues and perform actions with human-like decision-making. Source: https://help.kustomer.com/en_us/understanding-ai-agents-for-customers-10-SJWtPjAtC Last updated: 2026-05-26T17:55:10.787Z **Note:** This article covers usage of AI for Customers 1.0. Kustomer has launched AI for Customers 2.0, powered by the Kustomer AI Reasoning Engine. To learn more, see [Creating AI automations](https://kustomer.kustomer.help/en_us/creating-ai-automations-SyuRCs3tZl). AI Agents automate customer conversations and perform actions in Kustomer. They use artificial intelligence to respond to customer questions, complete tasks, and escalate issues when needed. AI Agents work alongside human agents to improve resolution speed and consistency. ## Key features * 24/7 automated support. * Multilingual conversations. * Multi-channel support (chat, email, SMS, WhatsApp, Facebook Messenger, and forms). * Tool-based actions such as updating orders or retrieving customer data. * Built-in guardrails and observability. **Who can access this feature?** **User types** Admins can access the AI Agent Studio page. ### In this article: * [What powers AI Agents?](#power) * [AI Agent teams](#teams) * [AI Agents](#agents) * [Supervisor agent](#supervisor) * [Tools](#tools) * [Knowledge](#knowledge) * [Observability](#observability) * [Guardrails](#guardrails) * [Reporting](#report) * [Best practice for disclosure of an AI bot](#disclosure) This article introduces the main concepts you need to understand and build an AI Agent team for your organization.  ### What powers AI Agents? AI Agents use a reasoning engine to evaluate customer input and determine next steps. The engine adapts to new information during a conversation and updates its response based on context. See this article on [AI models used by Kustomer](https://help.kustomer.com/en_us/ai-models-used-by-kustomer-SJNbqs5Lbe) for more details.  Agents generate responses dynamically and can take action using configured tools. ### AI Agent teams AI Agent teams group agents based on purpose or use case. There is no single AI agent to govern all use cases. Consider each agent as the authority over its own domain, thus a refund agent should govern refunds while a shipping agent should govern changes to shipping status, etc.  You should create teams based on their role, function, or channel: * An appointment scheduling team that supports booking, rescheduling, and preparing customers/patients for appointments.  * A team that handles "Where is my order" style questions. * A contract renewals team that handles outreach, co-terming, and approvals routing.  ![](https://cdn.kustomerhostedcontent.com/media/570fad9d9001bc1000163b28/44459c88c69423087688a9594c897bde.png) Each team should utilize its own tone, tools, and knowledge sources. When planning a team, define: 1. The problems the team solves. 2. The channels it supports. 3. The tools it can access. 4. The knowledge sources it uses. To create an AI Agent team: 1. From the left side navigation, click **Kustomer AI**. The **AI Automations** screen opens.  2. From the left navigation, click **Customers v1.0**.  3. At the top right side of the screen, click **Add AI Agent Team**.  4. Create your own **Custom Agent Team**, or **Browse Examples**. You can also interact with the **AI Agent Team Assistant** to build your assistant.  1. Examples are broken down by industry, designed to speed up agent creation for common issues. Click an example to prompt the system and begin the team creation process.  2. Custom Agent Teams give you full control over: 1. Greetings by channel 2. Usage of email templates 3. Ability to handoff to human agents 4. Customer identity verification 5. Fail messaging 6. Business scheduling 7. Routing behavior 3. Talking with the assistant will build these elements naturally in a conversational manner.  ![](https://cdn.kustomerhostedcontent.com/media/570fad9d9001bc1000163b28/5ca682a572efc9dc51cf1b84445587d1.png) ## How AI Agents work AI Agents use Large Language Models (LLMs) to interpret customer questions and determine actions. They operate in an iterative cycle: 1. Receive customer input. 2. Evaluate context. 3. Use tools if needed. 4. Generate a response. 5. Repeat as new information arrives. Agents can: * Retrieve customer details. * Update records. * Interact with external systems.  * Escalate to a human agent. When you configure an AI Agent, you define: * Its role. * Its response style. * The tools it can use. * The knowledge it can reference. Example: An airline booking agent might access reservation systems to change bookings or process cancellations. The airline check-in agent would handle check-in, boarding, and gate concerns. Each agent has access to its own necessary tools to route the conversation effectively.  ## Supervisor agent Each AI Agent team includes a supervisor agent by default. Customers interact only with the supervisor agent, even when other agents assist behind the scenes. The supervisor agent: * Greets customers. * Communicates directly with customers. * Accesses tools and knowledge sources. * Delegates tasks to other AI Agents on the team. You can define specific guardrails, give your agent guidance on responding to customer queries about your competitors, and more.  To edit the supervisor: 1. From the AI Agents for Customers page, click the team you wish to customize. The team's agent opens.  2. Click the supervisor agent to edit the agent.  Admins will be able to edit: * Description of the agent * Tools the agent can access * Knowledge sources the agent can access * The agent's instructions #### Tools Tools allow AI Agents to take specific action, effectively governing the actions an agent can take. Grant access carefully to ensure agents only perform appropriate actions.  Examples of tools include: * [Default tools to handle conversation routing](https://help.kustomer.com/en_us/default-tools-r1_tVKl61e) * Tools from applications connected through Kustomer * KObjects for customer attributes * Order history * Inventory * Specific functions (Get Refund request, Approve or deny refund request, etc.) See [this eCommerce example](https://help.kustomer.com/e-commerce-example-BJtDzOxJl) for sample use cases. ![](https://cdn.kustomerhostedcontent.com/media/570fad9d9001bc1000163b28/44fdcea53da1ec8d5f6621d635b2a41e.png) This [tutorial](https://help.kustomer.com/e-commerce-example-BJtDzOxJl) provides examples of AI Agents and tools you can consider using to respond to e-commerce inquiries. ### Knowledge AI Agents can reference your public knowledge base articles to answer questions. When searching for relevant information, agents use: * Article title. * Article body content. ![](https://cdn.kustomerhostedcontent.com/media/570fad9d9001bc1000163b28/507f952928ef5ce8b1f54655a27ac921.png) Agents do not use tags or categories when retrieving knowledge. #### Instruction Instruction guides your AI agent on its interaction with your customer. Think of this space as the agent's instruction manual. Writing simple and effective guidelines greatly improves your agent's interactions with your end users.  Basic instructions clarify: * The agent's identity. * Operating procedures (Step by step instructions that detail each action the agent should take).  * Other important guidelines not already covered. ### Best practices for disclosing an AI Agent When required by applicable law and to provide transparency to customers, it should be made clear that customers aren't interacting with a human and that the person is interacting with an AI agent.  Where possible, configure the agent to represent itself as a agent and not to obscure interactions as human-led. Examples of good configurations include: Goal Instruction Configure a customer support agent. "You are an AI customer support agent…" Configure a technical support agent. "You are an AI technical support agent…" It must be made clear to customers that they are communicating with an AI agent: * At the beginning of any conversation or message thread. * When a conversation moves from human interaction to automated experience. Acceptable disclosures include but are not limited to:  * “Hello! Welcome to \[Company name\]. I am an AI agent and am not a real person, but I’ll do my best to answer your questions.” * “Hello! I’m the \[Company Name\] AI agent,” * “Hello! You are interacting with an AI agent,”  * “Hello! You are talking to an AI agent,”  Set these disclosures by channel (chat, SMS, and email). **Note:** Even where not legally required, we recommend informing users when they’re interacting with an AI agent as best practice. Doing so helps manage user expectations about their interaction with your messaging experience. We recommend that you consult with your legal counsel to ensure that you are complying with all applicable laws regarding the use of AI Agents. ### **Prevent AI Agent re-engagement loops** When an AI Agent calls the RouteConversation tool and no human agents are available in the target queue, the routing system may reassign the conversation back to the AI Agent. This causes the AI Agent to run again in a loop, sending duplicate messages even after a human agent picks up the conversation. To prevent this behavior, add a guard condition to your routing workflow: 1. From the left navigation, select ****Settings****. The Settings page opens. 2. In the ****Settings**** page, select ****Platform****, then select ****Workflows****. 3. Open the workflow that handles AI Agent routing. 4. Add a condition step that checks: `assistant.status` does not equal `transferred` When an AI Agent calls RouteConversation, the system sets `assistant.status`to `transferred` on the conversation. Checking this value prevents the routing workflow from reassigning the conversation back to the AI Agent. ### **Configure transfer behavior** To ensure your AI Agent transfers conversations instead of closing them: * Remove the ****EndConversation**** tool from your AI Agent configuration. Use ****RouteConversation**** instead. * In your AI Agent supervisor instructions, explicitly state that the agent must not close a conversation when a routing or escalation tool has already been executed. * After making changes to an AI Agent or assistant configuration, republish the assistant for changes to take effect. ## Observability The **Traces** page logs AI Agent interactions. Traces help you: * Review conversation steps. * Understand tool usage. * Evaluate responses. * Identify areas for improvement. ![](https://cdn.kustomerhostedcontent.com/media/570fad9d9001bc1000163b28/f73846f812fdb294f95ca20cb315f0ff.png) Each trace includes: * Timestamps. * Customer inputs. * Tools used. * Responses generated. * Knowledge articles referenced, if applicable. See the [Observability article](https://help.kustomer.com/understand-observability-in-ai-agents-B1CMeXm1ye) to learn how to interpret traces. Learn more about observability and how to read a trace [in this article](https://help.kustomer.com/understand-observability-in-ai-agents-B1CMeXm1ye). ## Guardrails Guardrails define boundaries for AI Agent behavior. Use guardrails to: * Restrict sensitive topics. * Prevent competitor mentions. * Limit disclosure of confidential information. * Reinforce brand positioning. Guardrails work with built-in product safeguards to reduce policy violations. ## Reporting The AI Agents standard report tracks: * Conversation volume. * Response quality metrics. * Performance trends. Use this report to evaluate effectiveness and identify optimization opportunities. ![](https://cdn.kustomerhostedcontent.com/media/570fad9d9001bc1000163b28/cdea8d1832b4d5771677d509a6709dda.png)