Choosing a style for your agent

Default style

Use the Default style for simple, flexible workflows that rely on the LLM’s prompt-driven decision making. This style is tool-centric and allows the agent to decide which tool to use, when to use it, and how to respond without requiring a predefined plan.

When to use

Choose the Default style when your agent handles:

• Single step or lightly sequential tasks

• Scenarios that require adaptability

• Tasks involving multiple tools without strict order

Behavior

The agent uses iterative prompting to:

• Identify the appropriate tool or collaborator

• Determine the correct input

• Decide whether to continue or finalize the response

This loop continues until the agent gathers enough context to complete the task.

Tool compatibility

• Python tools

• OpenAPI tools

• MCP tools

Sample use cases

• Retrieve a report from a system

• Check the status of a ticket

• Extract information from a webpage

Benefits: Enables fast execution with minimal setup. Ideal for agents that need flexibility and quick responses.

ReAct style

Use the ReAct style for tasks that require step-by-step reasoning and validation. This style follows a "Think → Act → Observe” loop, allowing the agent to reflect on each action before proceeding. It is based on the ReAct methodology and supports transparent reasoning.

When to use

Choose the ReAct style when your agent handles:

• Exploratory or research intensive tasks

• Scenarios with evolving or ambiguous requirements

• Tasks that benefit from visible reasoning and user confirmation

Behavior

The agent performs the following steps:

• Think: Analyze the request and decide on the next action

• Act: Run the selected tool or collaborator

• Observe: Review the result and adjust the next step

This loop repeats until the task is complete.

Tool compatibility

• Knowledge intensive tools

• Data intensive tools

• Collaborator agents

Sample use cases

• Generate and refine code snippets

• Answer complex questions by using web search and synthesis

• Handle support tickets with multiple user interactions

Benefits: Improves accuracy through iterative reasoning. Ideal for tasks that require validation, transparency, or user feedback.

Plan-Act style

Use the Plan-Act style for structured, multi-step workflows. This style begins with a planning phase where the agent creates a sequence of actions. It then runs each step in order, by using predefined tools and collaborators. The agent can replan if conditions change during execution.

Plan-Act style requires setup by using the Agent Development Kit (ADK). For more information on configuring the style of your agent by using ADK, see Agent styles .

When to use

Choose the Plan-Act style when your agent handles:

• Business processes with multiple steps

• Tasks that require traceability and structured output

• Automations involving several tools or domains

Behavior

The agent performs the following:

• Generates a structured plan by using the LLM

• Runs each step in the plan sequentially

• Replans if unexpected results occur

• Summarizes the completed tasks (unless a custom output is defined)

Tool Compatibility

• Python tools

• OpenAPI tools

• MCP tools

• Collaborator agents

Sample use cases

• Create structured reports from multiple data sources

• Draft contracts or compliance checklists

• Combine results from search, calculator, and code tools

Benefits: Enhances reliability and traceability. Ideal for agents that require oversight and structured execution.

Recommendations

• Use Default for fast, flexible tasks with minimal dependencies.

• Use ReAct when accuracy and reasoning transparency are important.

• Use Plan-Act for structured workflows that require planning and traceability.