# Running Multiple Agents At Once With Warp

{% embed url="<https://youtu.be/w0bJFC0u0pE?si=XdP88cSCMXTK6Ec5>" %}

### 1. Why Multiple Agents Matter

Sometimes you need to work on several coding tasks at once — fix a PR, add a feature, debug a build — without losing context.\
\
Warp lets you run multiple agent tasks simultaneously, all within one workspace.

***

### 2. How It Works

Each agent runs in its own thread, complete with:

* Progress tracking
* Notifications when blocked or completed
* Separate command histories

Because Warp is a desktop app, it can send system notifications to alert you when an agent finishes or when it needs review.

***

### 3. Example: Reverting a PR and Editing a Shortcut

Ben uses voice mode to quickly start tasks.

Prompt Example:\
“Find the PR where we added the keyboard shortcut to the UDI input and revert it.”

He pastes in the PR number, and the agent:

* Locates the relevant diff
* Reverts the change automatically
* Pushes it to the correct branch

Warp notifies him when the task completes.

Then, he runs another prompt:

Prompt Example:\
“Change the keyboard shortcut to `Cmd + Shift + I`.”

Warp modifies `input.rs`, previews the diff, and Ben applies the change directly from Warp.

***

### 4. Managing Multiple Tasks

You can switch between concurrent agents:

* Each task appears in a Task List panel
* Completed, canceled, and running tasks are color-coded
* Toast notifications appear when tasks are blocked

You can even fast-forward agents to auto-approve all code diffs once you trust their trajectory.

***

### 5. Parallel Contexts

In another repo, Ben adds a new Eval test via a different agent:

Prompt Example:\
“Create a Python hello world function and verify it prints ‘Hello World.’”

Warp’s second agent:

* Locates the correct file
* Writes the test code
* Verifies execution

Meanwhile, the first agent continues working on the keyboard shortcut task.

***

### 6. Reviewing All Active Agents

Open the Agent Mode Dashboard to see:

* Active tasks
* Completed tasks
* Logs and outputs

You can refine or cancel tasks mid-run if needed, or switch back to manual commands.