Run tasks
Once you define a task and send it, the Air app runs it and you carry on working. What's left to you is where the agent runs, what you send it while it's busy, and how many tasks you keep going at the same time.
The Air app needs a Git repository in the project to create and run tasks. If the project doesn't have one, the app shows a warning and prompts you to initialize Git.
Choose where the task runs
A task can run in your workspace, in a separate Git worktree, in a Docker container, or in a cloud environment. The trade-off is isolation against startup speed: your workspace starts fastest and edits your project folder directly, while a container keeps every edit, command, and dependency away from your machine.
You pick the environment in the task header before you start the task, so it's worth deciding up front. See Run environments.
Work while the agent works
You don't have to wait for the agent between prompts. Send the next one and it joins the prompt queue, which dispatches prompts in order as the agent finishes the work in front of them.
Keep an eye on the context window indicator next to the chat input while a task grows. Once a task is using most of its window, it's usually cheaper to start a new one than to keep adding to it.
You can also run several tasks at once and switch between them when one needs your input, in one project or across all the projects you have open. See Multitasking.
When the agent is done, the result is waiting in your working copy rather than committed – see Review and integrate.