· 4 min
From seed to build sprint
A promising seed is not yet a product plan. Move into a build sprint when the experiment has narrowed the user, job and operating path enough that a small team can build something testable without hiding the remaining uncertainty. The sprint should answer the next product question, not reward a prototype for looking polished.
Check for evidence in four areas
- User pull: the people who face the job can describe when they need the result and how it fits their current process.
- Workflow fit: the required inputs can be obtained, reviewed and returned in a way that the user accepts.
- Quality boundary: the team knows which errors require a human, which cases should fail safely and what a useful output looks like.
- Operating cost: time, model spend, support effort and required integrations appear compatible with the product hypothesis.
Write down what is still unproven. Perhaps the workflow works on one document type but not another, or the user likes the output but has not committed to a recurring process. That uncertainty can become the sprint objective if the test is clear. It should not be disguised as an assumed product requirement.
Shape the first build sprint
Specify one user journey, one success measure, a limited set of inputs and the people who must review the result. Include access, data retention, error handling, cost visibility and a way to disable the model-assisted step. Agree on the scope boundary, dependencies and decision date before coding starts. Keep nonessential integrations and customization out until the first path is evaluated with users.
A positive gate may lead to a product sprint, a narrower experiment or a decision to stop. If evidence is mixed, choose the smallest next test that could change the investment decision. NIST’s voluntary AI Risk Management Framework is one reference for framing evaluation and risks across development and use; the framework does not determine whether this business should invest.
For the first use-case choice, see choosing an AI seed wedge; for pre-agreed stop rules, read kill criteria for AI experiments. To discuss a bounded build plan, visit services or contact.