Your AI agent can use coding helpers
When your AI agent writes code, it can now hand parts of the task to helpers that work in parallel, which makes bigger tasks faster and more thorough.
How to use it
- There is nothing to turn on. Your agent decides when a task is worth splitting
- You can ask for it in your prompt, for example by asking your agent to check several parts of the code at once
What you see
- Each helper shows as its own step inside the step where your agent works on code, labelled Looking into a sub-task, or Working on a sub-task (can make changes) when it can change files
- For now most helpers look into parts of the task and report back to your agent, which makes the changes. Helpers that can change files come with a newer setup for coding, and an agent that already has its setup keeps it, with the helpers it has, until that setup is renewed
- Each helper's own steps are listed under it, while it works and afterwards
Approvals
- With Ask first on, changes a helper wants to make wait for your approval too. The approval card says A helper asked, with the helper's task
- When someone you let prompt your agent, or a suggestion you accept, starts the work, a helper's changes always wait for your approval, whatever the setting
- Commands that only read something run without asking, as they do for your agent
Who sees helpers
- The same people who see your agent's work. If you hide the work, helpers' steps are hidden too
Who has it
Everyone whose AI agent can write code in their workspace: owners, admins and members by default, while a workspace's custom roles choose for themselves. Helpers' usage counts as the agent owner's, whoever prompted the work.
Good to know
- Helpers use more of your agent's usage, since each one works with AI models of its own
Where to find it
- In your agent's steps: open the step where it works on code, and its helpers are listed among the steps inside it