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

# Agents and Workflows: when to use which

> How Agents and Workflows differ, when to use each, and how to run a Workflow inside an Agent

Automate a task in Profound with an [Agent](/agents/about), a [Workflow](/agents/workflows/about-workflows), or both. Both live on the Agents page in Profound, both run on a schedule or on demand, and both produce Documents, Sheets, and other outputs. To see only one kind on the Agents page, use the **Type** filter.

## What's the difference?

Agents take written instructions, connected tools, and triggers, and AI Marketer (Aim) works out the steps on each run.

Workflows are automations you build yourself in the Workflow Builder as a chain of nodes that runs the same way every time.

The main difference is control. With Agents, you describe what you want to achieve and leave the implementation details to Aim. With Workflows, you build every step yourself and control the order of steps, the tools, and the inputs and outputs.

## When to use which?

Use an Agent for most tasks. If you're new to Agents, it helps you skip the implementation details and focus on the task you want your Agent to accomplish.

Use a Workflow when you need the same output every run, when a step must run code or call an API in a precise way, or when you need granular control over which steps run and in what order.

Workflows also tend to cost fewer credits for high-volume tasks. An Agent reasons through each step, and that reasoning uses credits.

If you need both, combine them: build an Agent and have it run a Workflow for the part that must stay fixed.

## Use a Workflow inside an Agent

An Agent can hand part of its run to a Workflow. The Agent handles the parts that need judgment, such as deciding what to look at, interpreting results, and writing, and the Workflow handles the parts that must run the same way every time.

### When to combine them

* **A step needs fixed logic.** For example, a calculation, a scoring model, code, or an API call that must run the same way on every run.
* **You already have a Workflow that works.** Reuse it as a step instead of rebuilding that part in the Agent's instructions.
* **You need an integration that Agents can't use as a tool yet.** Build a Workflow with that integration and have the Agent run it.
* **You want an Agent's triggers for an existing Workflow.** Start the Workflow from a monitor, an app event, or a Profound data change by running it from an Agent.

### Add a Workflow to an Agent

1. Make sure the Workflow is published. Only published Workflows can be added to an Agent.
2. Open the Agent and go to **Instructions** in the settings panel.
3. Place the cursor in the step where the Agent should run the Workflow, type **/**, and open **Agents**. Pick the Workflow from the list, or type its name to search.
4. The Workflow appears in the instructions as a pill with its name. Write around it what the Agent should give the Workflow and what to do with its results, for example, "For each topic from the input, run \[Workflow]. Then compare the results and write a summary."

Or ask Aim in the Agent's chat to add it, for example, "Use my Competitor Snapshot Workflow to pull the numbers."

To remove a Workflow from the Agent, hover over its pill in the instructions and select the **×**.

### What happens during a run

* When the Agent reaches the step, it starts a run of the Workflow and fills in the Workflow's inputs from the task.
* The Workflow's run shows in the Agent's chat with its status. Select **Open** to see the run in the Workflow, including what each node returned. The run also appears in the Workflow's own run history.
* If the Workflow includes a human review step, the run pauses until someone reviews the content.
* When the Workflow finishes, the Agent uses its results in the next steps.

### Requirements and limits

* **Access:** anyone who runs the Agent, including the person who set a trigger, needs access to the Workflow. When you share the Agent, share the Workflows it uses too.
* **Credits:** the Workflow's run uses credits the same way it does when you run it on its own.
* **Agents can't run other Agents.** Only Workflows can be added as steps.

## Side-by-side comparison

| | Agent | Workflow |
| - | - | - |
| **How you build it** | Describe the goal and the tools in plain English. Aim drafts the Agent, and you refine it in chat or edit the instructions. | Lay out each step as a node on a canvas, then configure and connect the nodes. |
| **How it runs** | Aim reasons through the steps on each run. | Follows the same fixed path every time. |
| **When the input doesn't match the expectation** | Works around it during the run and tells you what it did. | Stops or returns an error at the node that failed. |
| **Output** | Can differ between runs, within the guardrails you set. Wording and steps can differ between runs. | Identical structure every run. |
| **Seeing the work** | Every run is a chat: you see why each step happened, and the run is saved. | You see what each node returned. |
| **Tools and integrations** | Tools: connections to accounts in services like G2, Notion, Linear, or WordPress. Profound data is built in. | Integrations: nodes for each service, Profound data, code, and API calls. |
| **How it starts** | On demand, or on a Schedule, Monitor, Profound data, or app trigger. | On demand, on a schedule, or as a step in an Agent run. |
| **Using the other** | Can run published Workflows as steps. | Can't run Agents. |
| **API and Model Context Protocol (MCP)** | Not yet supported. Agents run in the Profound platform. | Supported through the Profound API and MCP. |
| **Credits per run** | Higher: reasoning on every step uses credits. | Lower for the same task. |
| **Best for** | Recurring tasks: monitoring, reporting with commentary, research, and drafting. | High-volume tasks that must be identical, code and API steps, and deterministic data operations. |