What's an agent in AI?

What’s an agent in AI?

One way to think of an agent is as a chain that knows how to make decisions. Where a chain follows a predetermined sequence of calls to different AI components, an agent uses a language model to determine which actions to take.

Agents are the part of AI that act as decision-makers. They can interact with other agents and tools. When you send a query to an agent, it tries to choose the best tools to use to answer. Agents adapt to your specific queries, as well as the prompts that configure their behavior.

sequenceDiagram
    participant User
    participant Agent
    participant LLM as Language Model
    participant Tool1 as Tool 1
    participant Tool2 as Tool 2

    User->>Agent: Submit Query
    Agent->>LLM: Analyze Query & Available Tools
    LLM-->>Agent: Recommend Tool 1
    Agent->>Tool1: Execute with Parameters
    Tool1-->>Agent: Return Results
    Agent->>LLM: Evaluate Results & Next Steps
    LLM-->>Agent: Recommend Tool 2
    Agent->>Tool2: Execute with Context
    Tool2-->>Agent: Return Final Data
    Agent->>LLM: Generate Response with All Context
    LLM-->>Agent: Formatted Response
    Agent-->>User: Deliver Complete Answer
```# Agents in `Agentic WorkFlow`

`Agentic WorkFlow` provides one Agent node, which can act as different types of agent depending on the settings you choose. Refer to the Agent node documentation for details on the available agent types.

When you execute a workflow containing an agent, the agent runs multiple times. For example, it may do an initial setup, followed by a run to call a tool, then another run to evaluate the tool response and respond to the user.