> ## Documentation Index
> Fetch the complete documentation index at: https://docs.workflowmachine.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Add steps

> Build the workflow sequence by adding app actions, AI steps, and control-flow steps in a clear order.

## Steps are where the workflow does the work

After the trigger starts the workflow, steps decide what happens next.

Each step should have one clear purpose. That might be:

* calling an app action
* transforming data
* generating text with AI
* making a decision
* waiting, stopping, or routing the flow

## Build the sequence one decision at a time

The easiest way to build a workflow is to ask:

`What should happen immediately after the trigger?`

Then ask the same question again for the next step.

This keeps the workflow understandable and avoids overbuilding too early.

## Common step types

Most workflows use a mix of these step patterns:

* **App steps** to create, update, send, or fetch something from another system
* **AI steps** to summarize, extract, classify, or draft content
* **Control-flow steps** such as **If**, **Wait**, and **End**
* **Data-handling steps** that prepare the output for the next part of the workflow

You do not need every type in every workflow. Use only what helps the workflow reach its outcome cleanly.

## Keep each step focused

A step is easier to test when it does one obvious job.

For example:

* one step classifies a message
* one step creates a record
* one step sends a notification

If one step is trying to solve too many things at once, it usually becomes harder to debug later.

## Add app actions carefully

When a step talks to another app, check:

* which connection it is using
* what fields are required
* what result the app step should produce

App steps often fail because of missing permissions, incorrect field mapping, or choosing the wrong saved connection.

## Add AI steps intentionally

AI steps work best when the task is narrow and the expected output is easy to evaluate.

Examples:

* summarize this content into three bullets
* classify this request as billing, support, or sales
* extract a few specific fields from this message

If the AI task is too open-ended, the workflow becomes harder to keep reliable.

## Use control-flow steps when the workflow needs decisions

Control-flow steps are useful when:

* one path should continue only if a condition is true
* the workflow should pause before the next action
* the workflow should stop early when the input is not useful

These steps make the workflow more flexible, but they should still stay easy to explain.

## Check the handoff between steps

Every time you add a step, confirm:

* what input it receives
* which previous step produced that input
* what output it should create next

Most workflow debugging comes down to bad handoffs between steps, not broken tools.

## Keep testing as you build

Do not wait until the whole workflow is finished before testing it.

Test after meaningful additions, especially when:

* a new app step is added
* an AI step changes the data shape
* a condition branches the workflow

Short feedback loops make workflow building much easier.