Agents

• UI agent - AI agent for web browser tasks. Used for dynamic and intelligent web automation. Simply write instructions to have it navigate websites, extract data, and generate structured outputs.

• Custom agent - AI agent for complex tasks. Create an agent that can understand instructions in natural language and take actions using available tools. Used for tasks that require reasoning, judgement, and dynamic planning.

UI agents

UI agent is a native agent that understands natural language instructions to perform complex browser actions. It can autonomously navigate websites, click, type, read data, and produce structured outputs optimized for downstream automation steps. Example use cases include summarizing products on a webpage or fetching data by navigating websites.

Properties

Title

Name of the step/UI agent

Instructions

In this field you write the prompt for the agent in natural language. Best practices while writing the prompt:

• Be clear and explicit about what you want.

• Structure the prompt. Start with mentioning the 'Task' or 'Role' first and then 'Instructions' to achieve the task with numbered steps

• Add constraints (e.g., only review the products section) and specify when to stop/end (e.g., stop when you find the relevant info)

• Provide positive and negative (don't do this) examples

• Specify length requirements (e.g., less that 100 words) or output format (e.g., date in MM/DD/YY format) clearly

Wrap the text in triple quotes (""") to write multiline prompts. For example:

"""Task: Locate the company's latest annual report.
* Visit the provided URL.
* Look for the annual report. The report may be titled 'Annual Report', 'Financial Report', 'Year in Review', or similar variations..."""

Structured Output (optional)

Agent Response: Name of the variable to assign the output of this operation

How to configure structured output fields

Adding fields

• Click Add field to create a new output field

• Enter the Output name - this becomes the JSON property name

• Select the Type from the dropdown

• Check Required if the field must always be present

• Add a Description to guide the AI agent

Field types

• String - Text values (Names, descriptions, summaries)

• Number - Numeric values (Counts, scores, percentages)

• Boolean - True/false values (Status flags, yes/no questions)

• Object - Nested structure (Complex data groupings)

• Array - List of items (Tags, categories, multiple values)

• File - File references (Document attachments, images)

• Data table - Tabular data (Structured datasets, reports)

Working with complex types

Objects and Arrays can contain nested fields:

• Click the expand arrow (▶) next to Object or Array fields

• Use Add field within the nested structure

• Keep nesting to 2-3 levels maximum for optimal performance

Example configuration

Here's a simple configuration for summarizing customer feedback:

{
  "orderId": "12345",
  "numberOfOrders": 3,
  "hasShipped": true,
  "orderDetails": {
    "quantity": 2,
    "productName": "ABC",
  },
  "tags": ["electronics", "urgent"]
}

This structure would be configured as:

• orderId (String, required)

• numberOfOrders (Number, required)

• hasShipped (Boolean, required)

• orderDetails (Object, required)

  • quantity (Number, required)

  • productName (String, required)

• tags (Array of Strings, Optional)

Best practices

• Use descriptive field names - Help the AI understand what data to extract

• Add clear descriptions - Provide context for complex fields

• Mark critical fields as required - Ensure essential data is always present

• Limit nesting depth - Keep structures simple for better performance

• Test your configuration - Verify the output matches your expectations by running the agent step and verifying the response.

Important notes

• JSON Knowledge: Unfamiliar with JSON? Learn the basics at json.org

• No validation: Currently, the system doesn't validate output structure - ensure your automation handles missing or malformed data

Custom agents

Custom agent is an intelligent action that processes natural language inputs to automate complex steps using integrated tool-calling capabilities. It primarily uses integrations as its tool interface, while offering extensibility to use Code as tool, and other native actions like human-in-the-loop task. The agent delivers structured, predictable outputs optimized for seamless integration into downstream automation steps.

Properties

Title

Name of the step/custom agent

Mode

A mode defines how the Agent operates based on your use case. The three available modes are: Fast, Pro, and Custom. Fast is best for simple tasks like summarization, classification, and high-volume automations, and Pro is ideal for complex tasks that involve reasoning and orchestration of multiple tools or actions. Fast and Pro are fully managed modes that require no pre-setup needed in advance. In Custom Mode, you'll need an Bedrock Converse connector and can select the model you want to use. This is ideal if you already have a prompt fine-tuned for a particular Bedrock model, specifically need a particular Bedrock model for the Agent, or want to include your own custom or fine-tuned model hosted on Bedrock. In Custom Mode, since you bring your own model from Bedrock via a connector, model inference is billed separately to the account associated with that Bedrock connector.

Instructions

In this field you write the prompt for the agent in natural language. Best practices while writing the prompt:

• Be clear and explicit about what you want.

• Structure the prompt. Start with mentioning the 'Task' or 'Role' first and then 'Instructions' to achieve the task with numbered steps

• To improve tool-call accuracy and guide the Agent, clearly specify in the prompt which tool to use at each step, if applicable.

• Specify length requirements (e.g., less that 100 words) or output format (e.g., date in MM/DD/YY format) clearly

Wrap the text in triple quotes (""") to write multiline prompts. For example:

"""You are content summarization agent.
Summarize the last two paragraphs of the provided text, focusing only on the main conclusion."""

Tools (Optional)

A tool is enables the AI agent to interact with external systems or perform specific tasks

General tools

Create user task

If enabled, this tool allows the Agent to trigger a Human-in-the-Loop (HITL) task whenever it gets stuck and needs assistance during execution. The Agent will pause and wait for human input, then resume once the required information is provided. The HITL task will be visible in the task center. For best results, the author can specify in the prompt exactly when the Agent should invoke HITL.

Integrations

If you've added specific connectors—such as Salesforce, MS Exchange, or Bedrock—to your automation group, their corresponding actions will appear here. The author can then select the relevant actions to use as tools for the Agent. For optimal performance, it's recommended to limit the Agent to 3–5 tools.

Structured Output (optional)

Configure your AI agent to return structured JSON output that can be easily processed in subsequent steps. This feature is ideal for text summarization, report generation, data transformation, and extracting statistics from unstructured content. This is an optional field. If you dont define structured output, the agent will by default return output in natural language.

Agent response: Name of the variable to assign the output of this operation

Note

The structured output configuration for Custom agents follows the same format as UI agents. Refer to the UI agent structured output section for detailed configuration instructions.

Custom agent testing

Users can test the agent independently of the full automation to validate behavior, debug prompts, and iterate faster.

Start test

• Hover on the agent card, a separate run button will show up on top of the card

• Click on the button to unit test this specific agent

• A variable collection window will pop up and automatically detect any variables used in the prompt/instruction

  • A preview of the prompt of this agent is displayed and highlights all the auto detected variables

  • Input put values for each variables before kick off unit test. Similar to all other expression fields of Amazon Quick Automate, the value of a given variable has to be valid expression syntax. Otherwise, an error will show up on the screen and prevent user to start test,

Test running

Users can see the execution log feed in the audit panel on the right side. The experience is the same as running the whole automation.

After test run

• User can see the input variables and output result at the Watch Variables` tab below the log feed.

• User can see basic metric card above the log feed (total time used and tools used).