Give Your Next AI Project a Definition of Done

By Mary Oge Chuks

“The content is ready” can mean several things. The words may be written while the image is missing, the links are unchecked or the publication date is wrong.

An AI-assisted project needs an explicit definition of done. Otherwise, the person handing over the work and the person receiving it may be describing different stages.

This is especially useful when a task contains many small steps. A blog post, a product listing or a client document can look complete before it is usable.

Describe the whole deliverable

Begin with what the recipient should actually receive.

For a blog post, that might include approved text, a featured image, accessible image description, categories, tags, a search description and the correct publication state. For a product, it might include the actual download and instructions as well as the sales page.

The list should fit the task. More checks are not automatically better. Include the details whose absence would cause a meaningful problem.

Separate preparation from completion

Use clear stage labels. “Written,” “ready for review,” “scheduled” and “published” are different states.

If a task requires an external action, the result of that action matters. A prepared upload is not a confirmed upload. A submitted request may still be processing.

This distinction protects everyone’s time. It prevents the recipient from assuming the final step has happened and discovering the gap later.

Decide what evidence is enough

Ask how completion will be demonstrated. A saved record, a working link or a successful readback may be appropriate, depending on the task.

Avoid demanding a large audit package for a small reversible change. The evidence should resolve a practical uncertainty.

Where a tool cannot perform part of the work, make that visible. “Everything except the sticky setting is saved” is useful. A broad “done” conceals the remaining action.

Review the definition after a real handover

When someone has to fix an omission, update the process. The lesson is not merely that one person forgot a step. It is that the task’s completion criteria failed to make that step explicit.

Keep the definition short enough to use again. A reusable checklist is valuable only if people can understand and follow it.

AI can help assemble complex work quickly. Clear completion criteria help turn that activity into a result someone can rely on.

Before the next project begins, finish this sentence: “We will call this done when…”