By Eric Tucker, Certified Scrum Trainer
What is a Definition of Done?
A Definition of Done describes the quality conditions an Increment must meet. It gives people a shared basis for understanding what is complete. The Scrum Guide requires teams to meet organizational standards as a minimum. Without such a standard, the Scrum Team creates one appropriate to its product. Developers must adhere to it; teams working on the same product share it.
That is different from a checklist an AI produces in isolation. A useful working question is: What evidence would let us confidently say this work meets our product’s quality standard?
Definition of Done vs acceptance criteria
Acceptance criteria describe expectations for a particular item. The Definition of Done supplies the shared quality standard. Use both when evaluating completed work.
| Comparison | Definition of Done | Acceptance criteria |
|---|---|---|
| Scope | Shared quality conditions for the product Increment | Expected behavior or conditions for a specific item |
| Example | Changed customer journeys pass the team's agreed accessibility checks | A customer can reschedule to an available time |
| What it answers | Does this work meet our quality standard? | Does this item do what we expect? |
For example, a rescheduling feature could satisfy its acceptance criteria but still fail the Definition of Done because its updated customer journey cannot be completed with a keyboard. The example is illustrative, not a universal checklist or a Scrum-prescribed accessibility test.
Start with a product example and the facts you know
We will use WorkshopBook, a fictional website where people book and reschedule workshops. Its team has seen changes that work on their own but break existing booking journeys. Their first draft says: “Code complete, tested, reviewed, and ready for release.”
That draft leaves too much open to interpretation. Tested how? Reviewed against what? Ready according to whom?
The prompts below are tool-agnostic and use invented product details. Replace them with your own context and approved, anonymized examples. Include organizational quality standards without pasting confidential documents into an unapproved tool. The examples are teaching aids, not evidence from a real customer or a complete quality standard for your product.
1. Ask AI for the missing context before drafting
Start with requirements, known problems, and how your team checks quality today. Giving AI a job title and asking for “the best Definition of Done” invites a generic answer.
Prompt to try
Help our Scrum Team develop a Definition of Done for WorkshopBook, a fictional workshop booking website. People book and reschedule workshops. Recent changes have broken existing booking journeys. Our current wording is “Code complete, tested, reviewed, and ready for release.” Before drafting, ask up to five questions about organizational standards, product risks, integration, user needs, and evidence of quality. Keep missing information visible. Do not invent policies or assume our tools.
Illustrative output: “Which existing customer journeys must keep working?” and “What organizational security and accessibility standards already apply?” are more useful starting questions than a list of fashionable engineering practices.
Team check: Answer what you can. Mark the rest as open questions with a person who will investigate. A gap in your information should not silently become an AI-generated requirement.
2. Generate a draft tied to evidence
Ask for a small set of clear conditions, not a long list of tasks. A condition should describe what must be true; evidence explains how the team knows.
Prompt to try
Using our answers and current Definition of Done, propose a draft with columns for quality condition, evidence, and unanswered question. Preserve every supplied organizational minimum. Cover integration, behavior, user access, applicable security checks, and information needed to use or support the changed product. Distinguish supplied requirements from suggestions. Do not invent numeric thresholds or present this draft as approved.
Illustrative output: “The change is integrated with the current product and the agreed booking and rescheduling checks pass” is more useful than “development finished.”
Team check: Ask whether each condition describes product quality or is simply a preferred activity. Running a test is an activity; the required behavior passing that test is evidence. Agree on what the check covers and where someone can find its result.
Example Definition of Done: a starting checklist
Here is a possible discussion draft for WorkshopBook. It needs the team's actual standards, supported environments, and evidence locations before adoption.
| Draft quality condition | Example evidence to agree on |
|---|---|
| The change is integrated and existing booking and rescheduling journeys still work | Passing results for the team's named regression checks against the integrated product |
| The changed behavior, including relevant error and recovery paths, meets the agreed expectations | Results linked to the item's acceptance criteria and applicable product checks |
| Changed customer journeys meet the applicable accessibility standard | Results of the required automated and manual checks, including keyboard use |
| The change meets the organization's applicable security and privacy requirements | Results and review records required by the existing policy |
| Customer-facing instructions and support information affected by the change are accurate | Updated information checked against the behavior of the integrated product |
| The resulting Increment is usable in the agreed supported environments | Demonstration and verification results for those environments |
This is deliberately a starting point. “Applicable” and “agreed” must point to named standards or checks your team can find. For example, replace “named regression checks” with the actual booking, cancellation, and rescheduling scenarios your team requires.
Do not add an arbitrary “100% coverage” target because AI suggested it. Choose checks that address your product's risks, and keep any mandatory organizational thresholds. A list that looks thorough but cannot be demonstrated will not settle disagreements about Done.
3. Replace vague phrases with observable conditions
Use AI as an editor to expose ambiguous language. Keep the team responsible for deciding which rewrite fits.
Prompt to try
Review this draft Definition of Done: [paste approved draft]. Identify phrases two people could interpret differently. For each, explain the ambiguity, suggest an observable replacement, and name the evidence we would need. If a threshold or policy is missing, ask us to supply it rather than choosing one. Do not remove an organizational requirement.
Illustrative output:
| Vague wording | More useful discussion draft |
|---|---|
| Tested | The agreed booking, cancellation, and rescheduling checks pass against the integrated change |
| Works well | The changed journey meets the team's named behavior and quality checks in its supported environments |
| Documentation done | Instructions affected by the change match the current behavior and are available where users or support staff need them |
Team check: Have two people independently explain what proof they would accept. If their answers differ, refine the condition or its evidence before calling it shared understanding.
4. Challenge the draft with failure scenarios
A helpful AI response can reveal questions your team has overlooked. It cannot prove the product is safe, accessible, or ready.
Prompt to try
Challenge our WorkshopBook draft using four hypothetical situations: an integrated change breaks an existing booking flow; a customer cannot reschedule using a keyboard; instructions describe the old behavior; and one required quality check has not run. For each, identify the relevant condition, whether evidence is sufficient to judge it, and any gap in the draft. Do not assume checks passed. Suggest questions for the team, not an approval decision.
Illustrative output: If keyboard access is required and the journey fails, the work does not satisfy that condition. If nobody has checked it, the status is unverified—not a pass.
Team check: Try the scenarios against your real draft. Avoid solving every scenario by adding another line; sometimes the existing condition simply needs clearer evidence. Ask the people doing the work whether the checks can happen as part of producing the Increment.
5. Facilitate the team's agreement
The Scrum Master can help the conversation stay focused on clarity and useful evidence. Involve the Scrum Team and the people who understand relevant product standards. Do not turn the exercise into the Scrum Master assigning a checklist to everyone else.
Prompt to try
Help me facilitate a 30-minute discussion of our draft Definition of Done. Propose an agenda that lets people identify ambiguous conditions, compare required evidence, surface disagreements, and preserve organizational minimums. Give me neutral questions rather than persuasion tactics. End with a way to distinguish agreed wording from unresolved questions. Do not equate silence with agreement.
An example meeting plan:
- Spend five minutes restating the product, known quality problems, and required standards.
- Spend ten minutes testing the draft against one recent piece of work.
- Spend ten minutes resolving wording and evidence disagreements.
- Spend five minutes recording the agreed wording and assigning follow-up for unresolved gaps.
This timebox is a facilitation suggestion, not a required Scrum event. A complicated disagreement may need another conversation. Keep unresolved items visible rather than declaring the draft adopted to finish on time.
Team check: Can everyone explain the same standard? Can Developers show how they will meet it? If several Scrum Teams contribute to this product, include them in the discussion rather than adopting incompatible standards separately.
Want to practice this conversation? Explore the live CSM course, including the in-class AI Definition of Done exercise.
6. Improve the standard using actual experience
After applying the Definition of Done, inspect where it helped and where it failed to make quality clear. Use AI to organize observations before your team considers changes.
Prompt to try
Compare our current Definition of Done with these approved, anonymized observations: [paste observations]. Separate unclear wording, missing evidence, conditions we did not meet, and possible gaps in the standard. For each proposed improvement, cite the observation behind it and explain the tradeoff. Preserve organizational minimums. Do not remove a quality condition just to make unfinished work appear Done.
Illustrative output: Repeated confusion over which environment was tested suggests naming the supported environments and evidence location. A failed required check calls for addressing the failure, not deleting the condition.
Team check: Discuss proposed changes in your Sprint Retrospective and assess how they affect future work and existing product quality. Keep a dated version so people know which standard applies.
What if the work does not meet the Definition of Done?
Be transparent about what remains incomplete. Under the Scrum Guide, an item that fails the Definition of Done returns to the Product Backlog and cannot be released or presented at the Sprint Review. That does not prevent discussing the problem and adapting plans with stakeholders.
AI cannot turn missing evidence into a pass. Keep “not checked,” “checked and failed,” and “checked and passed” distinct. Also distinguish meeting the quality standard from choosing when to release; the Sprint Review is not a required release approval gate.
Practice AI and Scrum together in a CSM class
Sustained Agility’s live CSM course with Eric Tucker, a Certified Scrum Trainer, covers Scrum, facilitation, coaching, and team effectiveness. The in-class AI exercise helps you create and improve a Definition of Done and evaluate what fits your team.
The prompts here are free examples of that approach, not a promise of an identical class script or a separate AI certification. You do not need AI experience to join.
See upcoming CSM class dates and prices. For the complete training and certification budget, read the CSM cost guide. If you are choosing between team coaching and product ownership, compare CSM vs CSPO. For product discovery and planning examples, explore AI for Product Owners.
Questions about AI and the Definition of Done
Can AI create a Definition of Done for my team?
AI can draft and critique one using your product context and requirements. Treat the output as discussion material. The team needs to resolve missing information, agree on clear conditions, and identify evidence before using it.
How is a Definition of Done different from acceptance criteria?
A Definition of Done sets shared product quality conditions. Acceptance criteria describe expectations for a particular item. For example, rescheduling behavior can meet its acceptance criteria while still failing a required accessibility check.
What should I put in an AI prompt for a Definition of Done?
Include your product, users, organizational standards, recent quality problems, current wording, and existing checks. Ask for observable conditions and evidence, and tell the tool to flag missing information instead of inventing requirements.
What is a useful first AI exercise for a Scrum Master?
Give the tool an approved draft Definition of Done and ask it to identify ambiguous phrases and missing evidence. Bring the proposed improvements to the team and test them against a recent piece of work.
Does Sustained Agility’s CSM class include this AI practice?
Yes. In the CSM class, you use AI to help create a Definition of Done and review what fits your team. The course focuses on Scrum and the Scrum Master role, with this concrete AI exercise included.