← Back to the journal

Adding background workers to my agent

An example build note about separating task acceptance from task execution.

The idea

Long-running agent work should not depend on keeping a web request alive. A small background worker is a useful place to begin exploring that boundary.

This is a starter example for a future build log, not a claim that this system has already been built.

The smallest useful flow

  1. Accept a task and persist it.
  2. Return a task ID to the caller.
  3. Let a worker claim and execute the task.
  4. Store the result so the caller can check its status.

What to capture in a real build log

Show the task lifecycle, a trace of one execution, and how the UI learns that the task finished. Include a link to the code and an honest account of what remains unreliable.

The next question

What happens when a worker disappears after claiming a task? That question leads directly into leases, timeouts, and retry behavior.