AI agent team guide
Plan and Test a Product Prototype With an AI Team
A falsifiable prototype task, test plan and evidence-to-revision log

An AI prototype team should make a product question easier to test. Start with the decision you need to make, choose the interaction that exposes the uncertainty and build only enough representation to learn. The useful return is a clear prototype brief and observation plan, not a polished mockup mistaken for validated demand.
Prototype Planner leads the experiment brief, Design Critic reviews the supplied flow, User Research Planner shapes the session and Feature Spec Writer records constraints and acceptance questions. The agents help plan and review; actual prototype creation and research use the tools and permissions your team controls.
Build your four-agent team
| Agent template | Its job on the team | Bring to its setup | Reviewable return | Team |
|---|---|---|---|---|
Prototype PlannerLead agent | Define the prototype’s learning job. | Decision, user task, uncertainty and known constraints. | An interaction map and bounded prototype brief. | Add to team |
Design CriticSupporting agent | Review the supplied flow against the task. | Actual screens or descriptions and the intended user action. | Ranked usability concerns with concrete revisions to consider. | Add to team |
User Research PlannerSupporting agent | Prepare a non-leading observation plan. | The decision, target participants and feasible research format. | Tasks, neutral questions and a synthesis plan. | Add to team |
Feature Spec WriterSupporting agent | Keep constraints and observable acceptance explicit. | Accepted scope, non-goals and relevant requirements. | A traceable brief that separates prototype behavior from production commitments. | Add to team |
Set up your team in Finta
- Add each template individually. Use the Add to team action for each named coworker, review the import setup and complete its Getting Started skill.
- Give each role the right context. State the shared objective, then supply the authorized documents, notes or exports that role needs. Do not assume an unconnected system is available.
- Review the handoff. Accept or correct one return before giving its evidence and open questions to the next coworker. Importing four templates does not itself run a bulk team.
- Make recurrence deliberate. Complete a useful manual pass first. Review any supported routine and enable it only when the scope, inputs and permissions are right.
Prepare the evidence
- A specific decision and the uncertainty the prototype can address.
- The actual target task and relevant current behavior.
- Available screens, interaction descriptions and known technical constraints.
- A research owner and a feasible, consent-aware plan for actual sessions.
Run the workflow
1. Choose a question the prototype can answer
Prototype Planner should name the decision before proposing screens. “Will users understand which item they are sharing?” is testable in a focused flow. “Will this business succeed?” is not a claim a small prototype session can establish.
2. Use the minimum meaningful fidelity
Choose enough detail to test the intended interaction. If wording is the uncertainty, preserve the real words. If navigation is the uncertainty, represent the choices. Feature Spec Writer records what is simulated and what is outside scope.
3. Critique the task path, not visual taste alone
Design Critic ranks concrete obstacles in the supplied material. Preserve the participant’s actual objective. The research plan should not explain the answer before the participant tries the task.
4. Separate observation from the decision
User Research Planner prepares how to capture behavior, questions and uncertainty. After real sessions, compare observations with the original question. Keep small-sample limits and counterexamples; the owner decides whether to revise, investigate or move forward.
A fictional prototype brief for a sharing decision
This example describes an invented interface and proposed study. No participant results have been collected.
| Brief field | Illustrative choice | Reason |
|---|---|---|
| Decision | Can a reader tell which material a share action includes? | Focus on one consequential comprehension question. |
| Representation | A short flow with the real proposed labels and a review state. | Represent the choices needed for the task, not the whole application. |
| Neutral task | Show what you would do to share only the selected material. | Avoid telling the participant which control to choose. |
| Observation | Record the selection, explanation and points of confusion. | Do not report findings before sessions occur. |
The right fidelity serves the question. A beautiful prototype with no learning objective can leave the product decision just as uncertain.
Copy the team brief
Prepare a prototype test brief for [specific product decision].
Use [user task], [current evidence], [supplied flow] and [constraints].
Prototype Planner: map the smallest interaction that exposes the uncertainty.
Design Critic: rank task-based obstacles and concrete revisions.
Research Planner: write neutral tasks, observations and synthesis questions.
Spec Writer: preserve non-goals, simulations and acceptance questions.
Return the brief, research plan, owner and decision boundary.
Do not invent participants or findings, build production software, recruit people or claim validation.Review the return
- The brief names a decision the prototype can genuinely inform.
- The represented fidelity matches the uncertainty and preserves material context.
- Observed evidence and the owner’s later decision remain separate.
Questions before you start
Does the template generate a working application?
This workflow prepares an interaction and test brief. It does not promise code generation, deployment or production reliability. Use the creation tools your team actually has for execution.
How many sessions prove the design works?
There is no universal number in this guide. Match the research design and coverage to the decision, and preserve what the evidence cannot establish.
Sources and next steps
This playbook uses the current published Finta marketplace packages linked in the team table. Open each profile to review its instructions, context, Skills and any routines before importing.