Building AI Agents

In deepset AI Platform, you can build AI agents using the Agent component incorporated into a pipeline. The Agent is a pipeline component that uses na LLM to call tools and reason about its actions. You can include multiple Agents in a pipeline and connect them to other components to build complex systems.

When to Use an Agent

Building an AI Agent makes sense for the following use cases:

• Multi-step tasks that require reasoning and decision-making. Cases when you can't predict the exact sequence of steps and the tools to use at each step. Examples:
  • Debugging code where the Agent needs to make hypotheses and try different solutions based on what it discovers.
• Tool orchestration where tasks require coordinating multiple tools or APIs. Examples:
  • An Agent deciding whether to query a database, search the web, or call an API based on the user's question.
• Dynamic planning when the path to the solution varies based on the user's input. Examples:
  • Customer support Agents that need to check the order status, process returns, or answer questions about the product based on the user's input.

When an Agent Might Be Too Complex

An Agent is a powerful tool but for some use cases it might be an overkill. Consider using a simpler alternative if:

• The workflow is predictable. You can map out the logic and steps in advance.
• Your application must be fast. Agents require multiple LLM calls which increases the response time.
• Your application must be predictable and highly reliable. Agents are less predictable and may be harder to control and debug.

For these use cases, a pipeline with tool calling or conditional routing may be a better fit. For approaches alternative to Agents, see Agentic Pipelines.