AI agents that use agentic AI are designed to solve complex user requests with limited direct human supervision. These AI agents are capable of autonomous decision-making, planning, and adaptive execution throughout the course of a conversation. This means they can interact with users in a more natural, human-like way, engaging in small talk and working through complex or vague issues by identifying the necessary next steps and asking follow-up questions where needed.
AI agents with agentic AI increase your automation rates and reduce the load on human agents and admins, while still providing the level of service your customers expect.
AI agents with agentic AI are available on messaging, email, and voice (EAP) channels.
This article contains the following topics:
Related article:
Benefits of AI agents with agentic AI
AI agents with agentic AI accelerate your path to 80% automation with minimal setup, full control, and built-in transparency. Powered by adaptive reasoning, these AI agents autonomously determine the most effective path to a resolution, handling nuanced queries across channels.
AI agents with agentic AI shift the paradigm of automated conversations. Instead of relying on rigid dialogue flows, they:
- Are purpose-built for service: Unlike other solutions that focus on broad enterprise workflows, these AI agents are designed for service to resolve even the most nuanced queries.
- Are easier to deploy and scale: Deploy these AI agents in minutes, not months—no complex flows, rigid templates, or manual training required.
- Provide better customer experiences: These AI agents don’t just execute tasks—they ask clarifying questions, determine the best actions, and adjust in real-time for natural, contextual interactions.
- Offer quality you control: Stay in command with full visibility into AI decision-making and the flexibility to let these AI agents operate autonomously or follow controlled workflows, ensuring every resolution meets your standards.
Ultimately, AI agents with agentic AI move past the limitations of scripted chatbots and first-wave generative bots, delivering higher automated resolution rates, lower operational costs, and a more engaging experience for end users.
Capabilities of AI agents with agentic AI
AI agents with agentic AI bring intelligence and flexibility to your support workflows by:
- Understanding incoming customer requests, routing them correctly, and resolving common issues automatically.
- Producing clear, customer-ready responses, aligned with your tone and brand.
- Escalating smoothly when needed, ensuring customer requests always progress toward a resolution.
The table below summarizes the main capabilities of AI agents with agentic AI.
| Capability | Description |
|---|---|
| Generative procedures |
Move away from fully scripted dialogues by simply describing your business policies and letting the AI agent create generative procedures that run the conversation autonomously. However, you still retain the ability to decide which topics should remain under scripted dialogues and which should use generative procedures. Generative procedures also allow for multilingual AI agents. You can write a procedure once, in a single language, and it can serve all the languages your AI agent supports. |
| Integrations and actions | The AI agent uses AI-driven procedures that can integrate your existing session and CRM actions, API calls, and more. All you need to do is to provide the AI agent with additional context in the form of an API description. |
| Adaptive reasoning | The AI agent analyzes each request from the user, identifies the issue, determines the best next actions, and adapts dynamically within the boundaries of your business procedures. It evaluates context, pinpoints missing information, and asks follow-up questions. Most powerfully, AI agents with agentic AI are capable of addressing multiple distinct issues within the same user request. |
| Reasoning controls | In the conversation logs, you can see the reasons behind an AI agent’s actions and decisions. Monitor how it solves problems, audit every conversation, and refine behaviors to ensure confidence in each resolution. |
| Disambiguation | When a customer's request is vague (for example, "I have a problem with my account"), the AI agent can ask for specifics before providing an answer or triggering a procedure. |
| Small talk |
For pure small talk with no question, the AI agent sends a response to proactively ask the customer for their query. For example: "Hi! I'm here to help with anything related to your orders. How can I help you today?" For small talk plus an in-scope question, the AI agent ignores the small talk and proceeds to resolve the customer's question. For small talk plus an out-of-scope question, the ticket is automatically escalated to a human agent. |
Differences in agentic AI capabilities on messaging and email channels
The table below describes how agentic AI handles features differently between messaging and email channels. Understanding these differences will help you set expectations and identify the best channel to use for a given AI agent.
| Feature | Agentic AI for messaging | Agentic AI for email |
| Generative procedures and dialogues | Use cases can trigger a generative procedure or a dialogue. | Use cases can trigger a generative procedure only. |
| Procedure execution | Multiple generative procedures can be executed sequentially during a single messaging conversation. | A single generative procedure can be executed per email response, but multiple procedures can be executed throughout the email chain (one per email). |
| Multi-topic support | Topics are handled one at a time in sequential, conversational turns. These topics can be questions that can be answered using connected knowledge sources (knowledge questions) or requests that trigger a procedure (procedure requests). | Emails with multiple knowledge questions are answered in a single email. Emails with a single procedure request are answered in a single email, and this process can repeat sequentially. Emails with multiple procedure requests, or a knowledge question and a procedure request, are escalated. |
| Responses | Multiple responses are sent, one to each topic as it’s addressed. | A single, unified reply is sent so customers receive a resolution in one message. |
| Parameter gathering | Parameters are collected over several turns, as questions are sent one by one. | All required parameters are collected in a single email, minimizing back-and-forths. |
| Escalations | Escalations are triggered by customer request, dialogues, or procedures. | Escalations are triggered by customer request, procedures, or any unsupported scenario, such as emails with multiple procedure requests, unsupported languages, or knowledge questions combined with procedure requests. |
| Automatic email cleanup | Not applicable | The AI agent automatically strips previous replies, signatures, footers, and legal notes from email threads, so it focuses only on the relevant new content in each customer email. |
Considerations and current limitations of AI agents with agentic AI
As you get started with AI agents with agentic AI, keep in mind the following considerations and current limitations:
- No support for rich-text formatting in generative procedures: The generative procedures used by these AI agents don’t support rich-text formatting (for example, buttons and carousels). However, rich-text formatting is supported in dialogues.
- No support for search rules in generative procedures: The generative procedures used by these AI agents don’t work with search rules for knowledge sources. That means you cannot have multiple knowledge sources and ask the AI agent to use only specific ones for certain users. However, search rules still work in dialogues.
- (Email channel only) Mixed question and procedure handling. The AI agent can’t automate responses to customer emails that combine both a question that can be answered with knowledge source content and a procedure request. For example, “Can I return my shoes? If yes, send me a label.” These types of emails are escalated to a human agent by default.
- (Email channel only) Parallel procedure automation. Only one procedure request can be automated per email response. The AI agent can’t respond to multiple procedure requests in a single email and will instead escalate these types of emails. The AI agent can, however, support sequential procedure requests. In other words, the AI agent completes one procedure, then the customer asks another question in a subsequent email that triggers a different procedure.
- (Email channel only) Limited control of email formatting. Control over greetings and the structure of the outgoing emails is limited. Future enhancements are planned to allow control over greetings, content layout, signatures, and footers. In the meantime, current personalization features and instructions do impact the AI agent’s email responses.
Next steps
See Getting started with AI agents - Advanced to learn how to create and configure an AI agent with agentic AI.