Why 'Can You Define Done?' Matters More Than Task Difficulty When Delegating to AI

A software developer argues that the key factor in successfully delegating tasks to coding agents is not task difficulty but whether clear acceptance criteria can be written before work begins. Tasks with mechanical, verifiable outcomes — such as passing tests or zero new lint warnings — hand off reliably, while vague tasks like 'improve performance' tend to produce repeated, unsatisfactory results. The author notes that this discipline of writing acceptance criteria first also benefits human team members, enabling colleagues unfamiliar with a codebase to take on bounded tasks with minimal context. A side effect of rigorous delegation is that only pure judgment calls remain on the developer's own plate, which proves more mentally taxing than expected. The author cautions that consistently delegating only the mechanical half of work risks hollowing out roles, and recommends retaining the decision-making step even when automation of it seems feasible.