What fills your week — and what could AI take off your plate?

Hypothesis

Most people can't list their own recurring work — but they can talk about it. A short, well-designed voice conversation should surface a person's Jobs To Be Done more completely than a form or workshop exercise, and an AI can score each job for automation potential while the person is still talking.

What we built

An eight-minute voice conversation with an AI coach. While you talk, the coach elicits your recurring jobs, scores each on frequency, pain, and AI feasibility, and captures the order you actually work in. You then review the map, correct anything it got wrong, and explore it as an interactive graph — your workflow in one view, AI-opportunity clusters in another, ranked by where AI could genuinely help first.

The coach's conversation design is a single carefully-built system prompt for the voice agent; the mapping and scoring happen live during the call rather than in post-processing.

Learnings

• Voice elicits far more than forms. People volunteer jobs they would never have typed into a text field — the conversational back-and-forth ("and what happens after that?") is doing real work.
• The review step is essential. The map is roughly right straight out of the conversation, but letting people correct it is what makes them trust it.
• Scoring during the conversation (not after) keeps the coach asking sharper follow-up questions about the painful, frequent jobs.

Log

• 2026-07-06 — Deployed to production at /experiments/jtbd-mapper.
• 2026-07-04 — Conversation design + agent prompt drafted; first end-to-end test calls.