The delegation test is whether you can write the check first
Every discussion I see about what to hand off to a coding agent sorts tasks by difficulty. Easy things to the agent, hard things to the human. It sounds obvious and it has been wrong in practice for me almost every time. The axis that actually predicts a good handoff is not difficulty. It is this: can you write down, before any work starts, what would make you accept the result? If you can write…
The true test of whether a task can be handed off to a coding agent or not is whether you can write down before starting the work, what would make you accept the result. If you can write the check, then the task can be delegated. If you cannot, then it is not ready for anyone to take on. Difficulty is not the deciding factor. A genuinely hard task with clear acceptance criteria has been completed successfully, while a trivial task that came back multiple times was due to an undefined outcome.
The key is to write the acceptance check first, as a separate document, before any instructions are written. This check should be checkable, such as cursor-based pagination on a specific endpoint, existing tests passing, and new tests for edge cases. Another example is upgrading dependencies across a monorepo with the build running smoothly and no new lint warnings.
An unspecified task, like making onboarding less intimidating, is not checkable as written. It involves a subjective judgement call on what constitutes intimidation, and what can be sacrificed. Once the judgement call is made, the task becomes checkable, such as reducing the signup form to three fields, deferring other validations, and maintaining existing validation behavior.
When all checkable tasks are delegated, the remaining work on your plate consists of purely subjective decisions, which are more mentally taxing than mechanical tasks. One surprising aspect of this approach is that rigorously delegating checkable tasks leaves behind only subjective judgment, which can be mentally exhausting. Another unexpected benefit is that such checkable tasks can be handed off to a teammate who has never worked on that part of the system before.
This reframe highlights that the check is not just agent infrastructure, but also people infrastructure. A clear acceptance criterion allows a teammate with no prior knowledge to pick up the task in just a few hours. This perspective led to writing acceptance criteria to empower agents, resulting in a more inclusive backlog where more team members can contribute. The check is the part you write, and it is the most crucial part of your job.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
