Get started with AI Agents
EARLY ACCESSTo use AI Agents, you need an active Infobip account with Chatbots enabled.
- If you do not have an account, sign up (opens in a new tab) for a free account. New users receive a 60-day free trial. For more information, see Create an account.
- To upgrade to a paying account, select Add funds at the top of the page. For more information, see Paying account.
AI agents fundamentals
An AI agent is an AI-driven system that can perform tasks autonomously to achieve goals. It uses large language models (LLMs), tools, external systems, and instructions to process requests and return responses.
Unlike traditional chatbots that follow scripted responses, AI agents can:
- Set sub-goals and plan multi-step workflows
- Use tools such as APIs, databases, browsers, and external services
- Take actions like sending API requests, updating databases, and retrieving information
- Identify the intents of the requests
- Adapt behavior based on context or previous interactions
How AI agents work
AI agents operate in a continuous cycle:
- Receive input: Accept requests from the solution in which they are used.
- Analyze and plan: Understand the goal, break it into steps, and identify which tools to use.
- Take action: Carry out the plan by calling APIs, querying databases, or using other tools.
- Evaluate results: Check if the goal was achieved or if additional steps are needed.
- Return results: Provide the final response or output.
This cycle may repeat multiple times for complex, multi-step tasks.
System architecture
AI agents must be used with a user-facing entry point such as a Chatbot:
- The Infobip solution receives a request from the end user and forwards it to the orchestrator agent.
- The orchestrator identifies which agents and tools can handle the request.
- Each agent or tool executes its task and returns a response.
- The orchestrator aggregates responses and returns the final result.
- The solution presents the result to the end user or processes it further.
Common use cases
- Recommend products: Help end users find products or services based on their preferences.
- Manage billing and payments: Process billing queries, payment updates, and refund requests.
- Track orders: Fetch real-time status updates by using tracking numbers or end user credentials.
- Answer FAQ: Answer complex questions by accessing internal databases, help centers, or third-party sources.
- Provide customer support: Resolve issues end-to-end.
- Book appointments: Schedule, reschedule, and cancel. Integrate with calendars.
Key terminology
| Term | Description |
|---|---|
| AI agent | An autonomous system that uses LLMs, tools, and instructions to complete tasks and achieve goals. |
| Orchestrator | An AI agent that coordinates subagents. Routes tasks, sequences steps, manages memory, and combines results. See Orchestration. |
| Subagent | A self-contained agent focused on a single goal. Plans, calls tools, and maintains local memory. Does not coordinate other agents. |
| Tools | External capabilities agents use to take action: Chatbot components, MCP servers, and Exchange integrations. |
| Component | A reusable Chatbot block that performs a task. See Component design. |
| MCP server | An external server that connects agents to services such as calendars, CRMs, or databases. See MCP servers. |
When to use AI agents
Not every use case requires AI agents. Use this section to decide whether to use AI agents or Chatbots.
Use chatbots when
- The use case is primarily FAQ or informational.
- The conversation follows predictable patterns.
- You need strict control over messaging for compliance, legal, or brand voice.
- Regulatory constraints require predictable behavior.
Use AI agents when
- The system needs to perform actions, not just provide information.
- Multiple backend systems need to be coordinated.
- The workflow involves conditional logic too complex for rule-based chatbots.
- Multi-step operations require dynamic decision making.
- The system needs to adapt to unexpected scenarios.
Comparison
| Chatbots | AI agents | |
|---|---|---|
| Interaction type | Primarily FAQ or informational | Performs actions, not just provides information |
| Conversation flow | Follows predictable patterns | Involves conditional logic too complex for rules |
| Control | Strict control over messaging for compliance or brand voice | Dynamic decision making across multi-step operations |
| Adaptability | Regulatory constraints require predictable behavior | System needs to adapt to unexpected scenarios |
| Integrations | Standalone or simple integrations | Multiple backend systems need to be coordinated |