Vivid Resume: the product decision that changed the whole system

The first idea was easy to describe—and too generic

“Use AI to create a better résumé” sounds useful. It is also broad enough to hide the important product questions. Better for which job? Based on what evidence? How should the user judge a claim? What else does someone need to apply with confidence?

A document-first product encourages a generation-first system: collect background information, prompt a model, return formatted text. That can produce an impressive demo. It does not necessarily improve the decision the user is trying to make.

The reframing: start with the target job

The product became more coherent when the target job moved to the beginning. That single decision established context for everything downstream: which experience matters, what language is relevant, where the evidence is thin, and what the application materials should help a reviewer understand.

The new product question became: How can AI help this person pursue this role while keeping every claim grounded in experience they can explain and defend?

This was not a copy change. It changed the information model, the workflow, the AI instructions, the review experience, and the boundary between assistance and invention.

Four parts of the system changed

1. Inputs gained a purpose

Work history stopped being a generic profile to summarize. Each role, achievement, and skill became evidence that could be compared with the target opportunity.

2. Generation became a transformation with constraints

The AI could tailor emphasis and language, but it needed to remain inside facts supplied by the user. Missing evidence became a product state to surface—not an invitation to invent.

3. Review became part of the core journey

The user needed to inspect, edit, and stand behind the result. Review was not a final polish screen; it was the point where control and credibility were preserved.

4. The output became a connected set of application tools

Once the job was the organizing context, the product could help with more than one document while keeping those outputs aligned to the same opportunity and evidence base.

A better quality bar than “the text sounds good”

Fluency is easy to notice and easy to overvalue. A useful evaluation plan needs representative jobs and user histories, then asks questions that reflect product risk:

• Is every material claim supported by user-provided evidence?
• Does the result emphasize experience relevant to the target role without erasing important context?
• Can the user understand why a suggestion was made and change it?
• Does the system handle thin, conflicting, or incomplete inputs without quietly fabricating?
• Do the application materials remain consistent with one another?

These checks are less exciting than a perfect demo. They are more important to a product people may rely on in a high-stakes moment.

The transferable lesson

The strongest AI product decision often happens before model selection or prompt design. Find the user decision, anchor the system around the context needed to support it, and decide what the product must never do.

Then make the first release prove that specific value. For Vivid Resume, the target job was not another input field. It was the organizing principle that made the rest of the product legible.