vibe-coding

Vibe Coding

Scope

Covers

• Timeboxed, AI-assisted rapid prototyping ("vibe coding") to produce a functional demo (not slides)
• Turning a rough idea into a buildable prototype spec + task board + prompt pack
• A tight iteration loop: generate → run → verify → adjust → log decisions
• "Build tools to build the thing" when it meaningfully speeds up the demo (timeboxed)
• Safe use of coding agents: least privilege, no secrets, small diffs, validation, rollback

When to use

• "Vibe code a working prototype we can demo in 30–90 minutes."
• "Replace this Figma concept with a clickable prototype."
• "I’m not an engineer—help me build a small app/tool with AI and ship a demo."
• "Turn this AI feature idea into a proof-of-concept with a clear build loop and demo script."

When NOT to use

• You need a production-grade system, hardening, scaling, or security review (use building-with-llms + engineering process).
• You need upstream problem framing, strategy, or PRD-level alignment (use problem-definition, writing-prds).
• The work is high-stakes/irreversible (payments, auth, medical, legal, safety-critical) without human owners and reviews.
• The request is "build anything" with no demo target; do intake first and narrow to one scenario.
• You want to evaluate whether to adopt a new AI tool or framework (use evaluating-new-technology).
• You need an AI/LLM product strategy, not a prototype (use ai-product-strategy).
• You're validating a startup idea at the business-model level, not building a demo (use startup-ideation).

Inputs

Minimum required

• Prototype goal: what should a user be able to do in the demo (1–3 "happy path" tasks)
• Target user + context (who uses it, where it fits)
• Timebox (e.g., 30/60/90 minutes) + demo audience (internal, customer, exec)
• Platform preference (web app, mobile, CLI, spreadsheet, etc.) and constraints (privacy, data sensitivity)
• Data/integrations: mock data ok? any APIs needed? (default to mock)

Missing-info strategy

• Ask up to 5 questions from references/INTAKE.md (3–5 at a time).
• If details remain missing, proceed with explicit assumptions and offer 2–3 options (e.g., mock vs real data; simple UI vs polished).
• If asked to run commands or write/modify files, request confirmation, keep changes in a dedicated folder, and include rollback guidance.