Working with Step Types

Every action is built from individual steps, and each step has a specific type that determines what it does. Understanding these types helps you build more sophisticated workflows and troubleshoot when things don’t work as expected.

Step Types Overview

Adopt AI provides twelve distinct step types, each designed for specific operations:

User Input: Collect information from users during workflow execution
API Call: Retrieve or send data to external systems
Data Processing: Transform, filter, or manipulate data
Intelligence Enrichment: Apply AI analysis to extract insights
Decision: Create branching workflows based on conditions
Master Plan: Define high-level workflow orchestration
Payload Generation: Construct structured request bodies
Metadata: Add contextual information to workflow execution
Story: Generate narrative explanations of workflow outcomes
Output: Format and present final results to users
Attribution: Track data sources and provide transparency
Recommendations: Suggest next actions to users

Let’s explore each type in detail.

User Input

These pause the workflow to ask the user for information. Use them when the initial request doesn’t contain everything you need.

When to Use

The user says “show me license renewals” but doesn’t specify a timeframe. Your action needs to ask “for which period?”

Example

Ask the user which timeframe to check: next 30 days, 60 days, 90 days, or custom range

Configuration Options

The question to ask
Input type (text, number, date, dropdown, multi-select)
Whether it’s required or optional
Default value if they don’t specify
Validation rules for the input

Best Practice: Keep user input steps to a minimum. If you can infer information from context, do so. Only ask when truly necessary, as excessive prompting can negatively impact user experience.

API Call

These connect to your systems to fetch or send data. Most actions include at least one API call.

When to Use

Whenever you need to get data from an external system, send updates, or trigger actions in other applications.

Example

Make a GET request to /api/licenses/upcoming with parameters:
startDate, endDate, status=active

Configuration Options

HTTP method (GET, POST, PUT, DELETE, PATCH)
Endpoint URL and parameters
Headers for authentication
Request body for POST/PUT requests
Which API integration to use
Timeout settings
Retry logic

Common Implementation Patterns

Fetching data: GET requests to retrieve information

GET /api/users/{userId}

Creating records: POST requests with the data to create

POST /api/tickets
Body: { "title": "...", "description": "..." }

Updating records: PUT or PATCH requests with changes

PUT /api/licenses/{licenseId}
Body: { "status": "active" }

Deleting records: DELETE requests (use with caution!)

DELETE /api/resources/{resourceId}

The response from an API call becomes available to subsequent steps. You’ll typically follow an API call with a data processing step to extract what you need.

Data Processing

These manipulate, filter, extract, or transform data from previous steps. They’re the workhorse of most workflows.

When to Use

After retrieving data that needs to be cleaned, filtered, sorted, or restructured before presenting to the user.

Common Data Processing Operations

Extracting fields:

Extract the "data" array from the API response

Filtering:

Keep only licenses where expirationDate is within the next 60 days

Sorting:

Sort by expirationDate ascending (soonest first)

Aggregating:

Calculate total renewal cost by summing the "annualCost" field

Transforming format:

Convert the date from ISO format to "MM/DD/YYYY"

Mapping:

Transform each record to include only: name, date, and cost

Rationale

Raw API responses often contain excessive or unstructured data. Data processing steps transform this information into clear, focused outputs suitable for end-user consumption.

Intelligence Enrichment

These use AI to analyze, interpret, or enhance your data with insights.

When to Use

When you need to extract meaning from unstructured data, classify information, generate summaries, or add AI-powered analysis.

Examples

Extracting structured data from text:

Parse the user's message to extract: application name, timeframe, and filters

Classification:

Categorize each support ticket as: Technical, Billing, or General

Summarization:

Generate a one-paragraph summary of these 50 customer reviews

Analysis:

Analyze sales data across all regions and identify the top 3 growth opportunities

Sentiment analysis:

Determine if this customer feedback is Positive, Neutral, or Negative

Pattern recognition:

Identify common themes across customer complaints

Strategic Value

These steps enable actions that transcend simple data retrieval by incorporating analytical capabilities and generating actionable insights. This intelligence layer differentiates sophisticated agent implementations from basic query systems.

Decision

These create branches in your workflow based on conditions. Think of them as “if/then” statements that determine the next steps.

When to Use

When different situations require different handling, or when you need to check conditions before proceeding.

Example

If the user's role is "Admin":
  - Proceed to show all licenses
Otherwise:
  - Show only licenses they own

Common Applications

Role-based access:

If user.role == "Manager":
  Include budget information
Else:
  Exclude budget details

Error handling:

If API call succeeded:
  Process the data
Else:
  Show error message and suggest retry

Data validation:

If startDate is after endDate:
  Return error: "Start date must be before end date"
Else:
  Continue with the query

Personalization:

If user's subscription tier is "Premium":
  Include advanced analytics
Else:
  Show basic metrics only

Configuration Options

Condition expression
True path (steps to execute when condition is met)
False path (steps to execute when condition is not met)
Multiple conditions (else-if logic)

Master Plan

These define the high-level orchestration strategy for complex workflows, coordinating multiple sub-workflows or parallel operations.

When to Use

For complex actions that require coordinating multiple independent workflows, parallel processing, or orchestrating several API calls that can run simultaneously.

Example

Master Plan:
Fetch user data (parallel with steps 2-3)
Fetch license data (parallel with steps 1 and 3)
Fetch billing data (parallel with steps 1-2)
Once all data retrieved, combine and process
Generate comprehensive report

Use Cases

Parallel data fetching:

Simultaneously retrieve data from multiple APIs to reduce total execution time

Multi-stage workflows:

Coordinate sequential phases where each phase may contain parallel operations

Complex orchestration:

Manage dependencies between different workflow branches

Strategic Value

Master Plan steps enable efficient execution of complex workflows by identifying opportunities for parallelization and coordinating multiple operations that would otherwise run sequentially.

Payload Generation

These construct structured request bodies or data payloads for API calls, transforming user input and processed data into the exact format required by external systems.

When to Use

When you need to create complex request bodies for API calls, especially when the payload structure is intricate or requires specific formatting.

Example

Generate API payload:
{
  "user": {
    "email": extracted_email,
    "role": user_role,
    "permissions": calculated_permissions
  },
  "metadata": {
    "created_at": current_timestamp,
    "created_by": current_user_id
  }
}

Common Use Cases

Creating complex POST requests:

Build nested JSON structure with user data, preferences, and metadata

Formatting batch operations:

Construct array of objects for bulk API operations

Dynamic payload construction:

Build payload conditionally based on user input and system state

Benefits

Ensures payload structure matches API requirements exactly
Handles data type conversions and formatting
Makes workflows more readable by separating payload construction from API calls
Enables reusability of payload templates

Metadata

These add contextual information to workflow execution, including timestamps, user information, execution context, and tracking data.

When to Use

When you need to enrich workflow data with contextual information, track execution details, or add audit information.

Example

Add metadata:
- Execution timestamp
- User ID who triggered the action
- Session information
- Request origin

Use Cases

Audit logging:

Track who performed what action and when

Execution context:

Capture environment, version, and configuration details

Data enrichment:

Add timestamps, user information, or system state to results

Analytics tracking:

Record metrics for workflow performance analysis

Strategic Value

Metadata steps ensure proper audit trails, enable troubleshooting, support compliance requirements, and provide data for analytics and optimization.

Story

These generate narrative explanations of workflow outcomes, translating technical results into natural language summaries that users can easily understand.

When to Use

When you want to provide users with a conversational explanation of what happened during workflow execution, especially for complex multi-step processes.

Example

Generate story:
"I checked your license renewals and found 12 licenses expiring in the next 60 days.
Three Adobe Creative Cloud licenses are the most urgent, expiring within 2 weeks.
The total renewal cost will be approximately $45,000. I've highlighted the urgent
renewals at the top of the list for your immediate attention."

Use Cases

Explaining complex results:

Translate multi-step analysis into clear narrative

Providing context:

Explain why certain results were returned or actions were taken

Guided interpretation:

Help users understand what the data means and what they should do next

Conversational responses:

Make agent interactions feel more natural and helpful

Benefits

Improves user understanding of workflow results
Creates more engaging, conversational experiences
Provides context that raw data cannot convey
Helps users know what to do with the information

Output

These structure the final result for presentation to the user—as tables, cards, lists, or narrative text. Output steps determine how information is visually presented.

When to Use

Almost always as one of the final steps, to ensure results are clear and easy to understand.

Formatting Options

Table format:

Format as a table with columns: License Name, Expiration Date, Cost, Owner
Sort by Expiration Date ascending

Card layout:

Display each renewal as a card showing:
- License name (bold header)
- Days until expiration (highlighted if < 30 days)
- Cost and owner details

List format:

Return as a bulleted list:
- License Name (expires MM/DD/YYYY) - $X,XXX/year

Narrative summary:

Generate a summary paragraph with key statistics and highlights

Mixed format:

Combine table for detailed data with summary statistics at the top

Configuration Options

Output format type (table, list, cards, text)
Columns to display (for tables)
Sorting and grouping rules
Highlighting and formatting rules
Conditional formatting based on values

Well-formatted output dramatically improves user experience. Choose formats that make information scannable and actionable.

Attribution

These track and display data sources, providing transparency about where information comes from and enabling users to verify accuracy.

When to Use

When aggregating data from multiple sources, when users need to verify information, or when compliance requires source tracking.

Example

Add attribution:
- License data from: License Management System API
- Cost data from: Finance Database
- Owner information from: HR System
- Last updated: 2024-02-19 14:30 UTC

Use Cases

Multi-source aggregation:

Track which data came from which system when combining multiple APIs

Compliance and audit:

Document data lineage for regulatory requirements

Transparency:

Show users where information originated

Verification:

Enable users to validate data by checking sources

Benefits

Builds user trust through transparency
Enables data verification and validation
Supports compliance and audit requirements
Helps troubleshoot data quality issues

Recommendations

These suggest what users might want to do next, creating a more proactive and helpful experience.

When to Use

At the end of an action to guide users on logical next steps or related actions they might find useful.

Example

Based on these upcoming renewals, suggest:
- "Would you like to see the budget impact?"
- "Should I notify the license owners?"
- "Do you want to review past renewal costs for comparison?"

Common Recommendation Patterns

Follow-up actions:

Suggest related queries based on current results

Workflow continuation:

Offer next logical steps in a business process

Feature discovery:

Introduce related capabilities users might not know about

Contextual help:

Suggest relevant documentation or support resources

Strategic Purpose

Recommendations maintain conversational continuity and facilitate capability discovery, improving overall user engagement and system utilization.

Combining Step Types

The power of actions comes from combining different step types into comprehensive workflows. A sophisticated action might include:

User Input - Collect parameters
Payload Generation - Construct API request
API Call - Fetch data from external system
Data Processing - Filter and transform results
Decision - Apply role-based or conditional logic
Intelligence Enrichment - Generate insights using AI
Metadata - Add execution context
Output - Format results for presentation
Story - Generate narrative explanation
Attribution - Document data sources
Recommendations - Suggest next steps

Next Steps

Now that you understand step types, you’re ready to:

Test actions with different step combinations
Build advanced workflows using multiple step types
Apply best practices for robust action design

​Step Types Overview

​User Input

​When to Use

​Example

​Configuration Options

​API Call

​When to Use

​Example

​Configuration Options

​Common Implementation Patterns

​Data Processing

​When to Use

​Common Data Processing Operations

​Rationale

​Intelligence Enrichment

​When to Use

​Examples

​Strategic Value

​Decision

​When to Use

​Example

​Common Applications

​Configuration Options

​Master Plan

​When to Use

​Example

​Use Cases

​Strategic Value

​Payload Generation

​When to Use

​Example

​Common Use Cases

​Benefits

​Metadata

​When to Use

​Example

​Use Cases

​Strategic Value

​Story

​When to Use

​Example

​Use Cases

​Benefits

​Output

​When to Use

​Formatting Options

​Configuration Options

​Attribution

​When to Use

​Example

​Use Cases

​Benefits

​Recommendations

​When to Use

​Example

​Common Recommendation Patterns

​Strategic Purpose

​Combining Step Types

​Next Steps

Step Types Overview

User Input

When to Use

Example

Configuration Options

API Call

When to Use

Example

Configuration Options

Common Implementation Patterns

Data Processing

When to Use

Common Data Processing Operations

Rationale

Intelligence Enrichment

When to Use

Examples

Strategic Value

Decision

When to Use

Example

Common Applications

Configuration Options

Master Plan

When to Use

Example

Use Cases

Strategic Value

Payload Generation

When to Use

Example

Common Use Cases

Benefits

Metadata

When to Use

Example

Use Cases

Strategic Value

Story

When to Use

Example

Use Cases

Benefits

Output

When to Use

Formatting Options

Configuration Options

Attribution

When to Use

Example

Use Cases

Benefits

Recommendations

When to Use

Example

Common Recommendation Patterns

Strategic Purpose

Combining Step Types

Next Steps