Supervised AI · Development workflows
Coding agents with their own workspace
I tested a supervised remote coding workflow with a separate working copy for each task and context that could be resumed. The useful core was demonstrated and preserved when the wider automation experiment was simplified. The worker was then paused.
Keep task state separate
Its own working copy
Changes and context stay with the task.
Another working copy
Work can continue without sharing mutable files.
A coding task can outlast a local editing session. Moving the work elsewhere creates another question: how can tasks keep their changes and context without sharing a mutable checkout?
I directed and tested a remote coding-agent experiment using separate Git worktrees and resumable task context. Implementation was AI-assisted, with decisions and review remaining part of the workflow.
The core experiment demonstrated repository work in an isolated remote workspace. During a later simplification, I kept the useful technique and paused the worker rather than continuing the larger platform ambition.
The experiment was preserved in July 2026. It does not establish a currently operating autonomous delivery service or a complete ticket-to-delivery product.