Skip to main content

AdaL Engineer

Experimental

AdaL Engineer is an experimental feature and is under active development. Expect rough edges — the default coding agent is the recommended way to work.

AdaL Engineer — supervisor delegating to parallel workers

adal --agent-mode engineer

Then describe the task in the prompt. The supervisor will plan, delegate, and report back as workers make progress.

For headless or scripted runs:

adal -q "refactor the auth middleware to use Clerk's session token" --agent-mode engineer
note

AdaL Engineer is currently available in the CLI only and is not supported in the Desktop app.

Why AdaL Engineer​

AdaL Engineer is becoming the default way to work with AdaL. It goes beyond traditional loop engineering by combining a supervisor-worker architecture with parallel delegation — giving you the depth of a senior engineering team at an affordable price.

The goal is simple: make loop engineering accessible to everyone. You describe the work, AdaL plans and executes it across parallel workers, and you review the results. No manual step-by-step prompting, no context switching between tasks.

The future of engineering is here: start your day by defining a few goals, assign them to AdaL Engineers, and shift your focus to the "why" — why this feature matters, what problem it solves for users — while AdaL handles the implementation. Engineers become product thinkers, not ticket executors.

Use it for things like:

  • Multi-file refactors across an unfamiliar codebase
  • Investigating a bug that spans services or components
  • Building a feature end-to-end (plan → implement → test → review)
  • Multi-tasking — running several independent tasks in parallel, each handled by its own worker
  • Any task where you want AdaL to organize the work, not just respond to one prompt

Safety Notes​

AdaL Engineer can run multiple workers in parallel and write to files across your project, so use it thoughtfully.

  • Use with caution. Workers may modify files outside the path you mention if the task requires it.
  • Prefer working in a git worktree or branch so reviews stay scoped.
  • Review the diff before merging — workers optimize for the task, not for minimal changes.
  • You are responsible for the actions AdaL Engineer takes on your behalf. AdaL is provided as-is, without warranty of any kind, and AdaL is not liable for any changes AdaL Engineer makes.

For browser-based debugging sessions, see Browser Use. If Browser Use doesn't work, check the Browser Use setup guide for OS-specific installation instructions.

Related: Agents & Modes (/agent) · Headless Mode · Browser Use · Slash Commands

Worktrees​

AdaL's worktree tool puts each piece of parallel work in its own isolated checkout — Engineer mode puts a worker in each and merges the results back.

For how worktrees work, the worktrees.toml config, cleanup semantics, and running AdaL in a worktree yourself with native git, see Worktrees.

Running AdaL in a worktree yourself​

The worktree tool is agent-only — there is no slash command for it. If you want to run AdaL in a worktree manually, use native git:

git worktree add .agents/worktrees/feature-auth -b worktree-feature-auth
cd .agents/worktrees/feature-auth && adal

git worktree list
git worktree remove .agents/worktrees/feature-auth

Running adal inside a worktree works exactly like running it anywhere else — a worktree is just a directory.