Interviews and Decisions
On this page
Interviews and Decisions
A short PM-style interview before each generation, and a read-only log of every decision it produces. · Optional · Pro plan
When guided mode is on and you're on a Pro plan, the agent asks a few clarifying questions just before a generation runs — and records your answers, along with the plans you accept or reject, into a decision log that surfaces on the project Summary and in your PRD export. This whole layer is optional: you can skip the questions on any run, and free-plan users simply don't see it.
The pre-flight interview
Before a pipeline generation kicks off in guided mode, the agent may ask up to three clarifying questions — the kind a product manager asks before writing a spec. Answering shapes the output; the interview uses the same model you've selected in the chat input's model picker.

Each question is a card with option pills (single- or multi-select) plus an always-available free-text box. A progress line reads "1 of 3 — answers shape this generation". When you've answered the last one, the generation proceeds with your answers folded into its context.
💡 Tip: Even a one-word answer helps. The questions target the ambiguities the generator would otherwise guess at — audience, scope, priority — so a quick reply meaningfully sharpens the result.
It never blocks you
The interview is a soft gate. A Skip — generate now button sits on every question, and clicking it runs the generation immediately. Anything you'd already answered still rides along, so skipping half-way keeps the value you added. The gate also fails open automatically: if questions can't be fetched, time out, or come back empty, the generation just proceeds as normal. A guided interview can never stand between you and a generation.
Plan gating
The interview is a Pro-plan feature (and above). The deterministic guided features in 9.1 — arrival briefs, reviews, and the coverage check — are free for everyone; only this LLM-powered interview layer is gated. Free-plan users on guided mode see a one-line upgrade hint instead of questions, and their generations run exactly as before. See 15.1 — Account and billing for plan details.
The decision log
Every answered interview is recorded as an immutable entry in your project's decision log, alongside the plan proposals you accept or reject when the agent asks for confirmation. Each entry captures the decision, its rationale (the question that prompted it), and where it came from.
Entries are tagged by source:
| Source label | Where it comes from |
|---|---|
| Interview | Answers to a pre-flight interview question |
| Agent plan | A plan proposal you accepted or rejected |
| What-if | Outcomes of what-if explorations |
The log is read-only — there's no add, edit, or delete. It's a record of what was decided and why, not a task list.
Where it surfaces
- Project Summary — a Decisions section on the 3.1 — Project Summary page lists the log newest-first, each with its source tag and date. The section is hidden entirely while the log is empty, so pre-guided projects see no change.
- PRD export — the same decisions appear as a
## Decisionssection in your exported PRD, so the reasoning travels with the spec to your coding agent.

⚡ Power user: Because decisions ship inside the PRD export, your downstream AI coding agent (Cursor, Claude Code, Windsurf, Copilot) receives not just what to build but the choices behind it — the context a human dev would otherwise have to re-ask for.
↔ The traditional way
On a team, these decisions live in meeting notes, Slack threads, and a PM's memory — and they evaporate by the time an engineer picks up the ticket, forcing a round of "why did we decide X?". The decision log makes that trail an automatic, durable by-product of building the spec, and carries it straight into the developer handoff.
What's next
- 3.1 — Project Summary — where the decision log surfaces.
- 10.1 — Prepare for Dev — turn the finished spec into a build-ready blueprint.