Define Done for an Agent-Built Deliverable

Shashank Agarwal

Shashank Agarwal

#engineering-leadership#founder#agent-evaluation

An agent can produce a draft, run a command or submit a request while the user's intended result remains unfinished. A useful definition of done identifies the actual artifact and its verified state.

Quick answer: Define what must exist, which conditions it must satisfy, how those conditions will be checked and where the user will receive it. Separate attempted actions from established outcomes. Record unresolved limits instead of describing a partially verified artifact as complete.

Name the deliverable

For an illustrative report task, “write the report” could mean a local draft, a saved shared document or a published page. The acceptance contract should identify the intended state.

RequirementExample verification
ContentRequired questions answered with appropriate evidence
ArtifactCorrect saved document or page
LinksRelevant destinations resolve
DeliveryIntended user can access the result
PublicationPublic page exists if publication was requested
LimitsMissing or unverified findings stated

This is a proposed completion checklist. It does not imply permission to publish when the user only requested a draft.

Test what matters to the consumer

A generated JSON file may parse while containing the wrong fields. A page may exist while its required link is missing. A successful save request may leave the editor showing an error.

Inspect the final artifact against the actual task, not merely the tool's success message.

Preserve partial progress honestly

If a service is unavailable, keep the prepared work and record what remains. A clear draft and exact blocker are useful outcomes, but they should not be counted as a live placement.

Make completion reviewable

Provide the concrete artifact and a compact account of what was verified. The user should not need to reconstruct success from a long tool log.

A definition of done is strongest when another reviewer can repeat the relevant checks and reach the same bounded conclusion.

I'm Shashank Agarwal, founder and CEO of Noveum AI and creator of API.market. See the acceptance-criteria guide and evaluation-report guide for ways to make a technical decision reviewable.

Follow my work on LinkedIn. Drafted with AI assistance.

Exclusive Early Access

Get Early Access to Noveum.ai Platform

Join the select group of AI teams optimizing their models with our data-driven platform. We're onboarding users in limited batches to ensure a premium experience.

Early access members receive premium onboarding support and influence our product roadmap. Limited spots available.