How to begin a small project with AI

When an idea appears, the first need is rarely a grand plan. A team needs a small and concrete outcome that lets everyone picture the same situation. AI can help produce that outcome quickly, but letting it decide the direction can blur the reason the project exists.

Narrow the problem to one sentence

A useful opening sentence describes change rather than a list of features.

Help a caregiver find a free neighborhood activity for this weekend in under three minutes.

This sentence includes an audience, a situation, and a desired change. “Build a local information platform” is too broad to reveal what should be tested first.

Three questions are enough at the beginning:

1. Who will use the first outcome?
2. What friction do they experience now?
3. What observable scene should change this week?

Once the answers are clear, requests to AI become easier to review.

Give AI work that people can verify

“Make a good service” is difficult to evaluate. Break the work into research, classification, drafting, comparison, and review.

• Find ten public examples and compare their common patterns.
• Draft seven questions for an initial user conversation.
• Compare three information structures for a single screen.
• Mark vague language and missing conditions in a draft.

People still make the consequential decisions. Ask AI to identify facts that need verification, sources used, and low-confidence areas so the review does not begin from zero.

Let the first outcome move the conversation forward

The first outcome can be modest: one clickable screen, a page filled with realistic copy, or a paper prototype. Its purpose is not to look finished. It should make better questions possible.

• Where does a person pause?
• Is the feature we assumed was essential actually needed?
• Can the flow be understood without an explanation?
• Are public and private information clearly separated?

If the team can answer those questions, the outcome has done its job.

Leave a record for the next task

Saving only the final result causes teams to repeat old mistakes. Keep a short note of the first assumption, why a choice was made, what changed, and what remains unknown. AI may organize the discussion, but people should verify every decision and source in the final record.

The same rule applies when a Haru Space pet drafts an article or artifact. The pet can propose, compare, and find omissions. Group members verify publication scope, facts, and rights before approving it. That boundary allows speed and trust to coexist.