# Build the smallest thing that can answer the important question.

> AI makes it inexpensive to demonstrate a possibility. That can be a gift or a trap. The right prototype answers a specific product or architecture question, rather than merely looking like a finished application.

Canonical URL: https://marmation.com/solutions/product-discovery-prototyping/

## The decision

What single uncertainty should this prototype resolve?

Choose the artifact that can answer that question: a low-fidelity interaction, a technical spike or an evaluation harness. Commit further only when the evidence warrants it.

## The situation behind the request

A product leader has a concept for an assistant that reviews complicated customer submissions. Sales sees opportunity; engineering sees uncertain data quality, latency and integration risk. A polished interface will not resolve those unknowns.

## Where the obvious solution falls short

Teams often confuse a demo with validated demand, technical feasibility or a business case. Prototypes can also hide the cost of API usage, support, permissions and messy input.

## How we would approach the work

We define a hypothesis, user, acceptance case and one hard unknown. Then create an interactive product slice, technical spike or evaluation harness designed to test that unknown with realistic or synthetic inputs.

## What the engagement may include

- Product opportunity and user workflow brief.
- Interaction prototype with uncertainty, correction and review states.
- Technical feasibility slice and option comparison.
- Evidence summary, decision criteria and production roadmap.

## The choices worth making explicitly

Use low-fidelity interface work before sophisticated engineering when usability is the main unknown. Use a narrow technical spike when model or integration feasibility is unknown. Do not build both for show.

## What we would test before calling it ready

Test actual tasks with representative users where permitted, validate input/output quality, estimate latency and cost, and identify unproven assumptions. Never equate positive reactions with purchase intent without evidence.

## A sensible first step

Name the one question a prototype must answer. A well-designed validation can tell the team not to build the feature, which may be the most valuable outcome.

## Related reading

- [Build AI into the product. Not just the prototype.](https://marmation.com/solutions/ai-product-engineering/)
- [Give an agent a job. Give the system boundaries.](https://marmation.com/solutions/ai-agents-and-workflows/)
- [A good AI decision is worth more than a rushed AI build.](https://marmation.com/solutions/ai-consulting-architecture/)