What is Build With Billy?

Build With Billy is a personal brand and community from Billy Lewis — a self-taught builder running AI systems across a handful of small businesses from a farm in New Zealand. It’s a free newsletter, a free monthly community call, practical guides, and a couple of products (the Brain Kit and a cohort) for people who want to build real things with AI rather than just chat with it.

How much does AI consulting cost in New Zealand?

AI consulting in NZ ranges from free community workshops to $297–$497 for an AI Opportunity Map diagnostic, to custom retainers for managed automation. Build With Billy offers a free monthly community call, a time-savings calculator, and paid consulting starting at $297 NZD.

Do I need technical skills to use AI in my business?

No. Most AI tools have free tiers and simple interfaces. Build With Billy’s clients typically start with $0/month tools like Claude, Airtable, Supabase, and Vercel. Everything is explained in plain English for regular people doing meaningful work.

Can a small NZ business afford AI automation?

Yes. Most AI tools have free tiers. Build With Billy’s clients typically start with $0/month tools and only add paid services when the ROI is proven. The AI Quick Wins checklist shows 5 automations you can set up for free today.

A living knowledge base that remembers every decision, insight, and pattern. It grows with your work and gives your AI permanent context. Available as a starter kit from Build With Billy.

Is the /interview skill really free?

Yes. It's free and MIT-licensed. Clone the repo, copy the skill into your Claude Code skills folder, and use it on a weekend hack or a client flagship.

Do I need Claude Code to use it?

No. The Claude Code track gives you auto-triggering and real enforcement, but there's also a chat track — a single copy-paste prompt that works in Claude, ChatGPT, or Gemini with no install.

What does it actually produce?

Four documents: constitution.md, spec.md, plan.md, and tasks.md. Paste them into a repo, hand them to a coding agent, or keep them as your source of truth.

FREE RESOURCE · OPEN SOURCE · MIT

Interview yourself
before you build.

/interview is a Claude Code skill that runs a London-ad-agency-depth discovery interview, drills every facet of what you're about to make, turns your answers into a complete spec — then keeps the build safe. Free. Installs in 60 seconds.

Install the skill How it works

Wraps GitHub Spec Kit · works for code and non-code projects

WHY IT EXISTS

Most projects don't fail in the build. They fail in the brief.

You ask one question, you build the wrong thing. The moment you start coding something you haven't fully thought through, every hour after that is at risk.

A senior agency lead doesn't take a brief at face value — they drill “and what does that actually mean?” until there's nothing left to abstract. /interview makes your AI do exactly that — then writes the spec, then keeps the build safe.

HOW IT WORKS

You don't answer questions one at a time.
It interviews you.

The first thing it asks: what are you building, and how heavy is it? From there it runs a structured discovery and ends with a complete spec.

Detect context + weight

Code or non-code? QUICK, STANDARD, or FLAGSHIP? The weight sets how deep it drills and how much it stops to check.

Pre-load what's known

It pulls existing context first, then confirms its assumptions instead of re-asking — so the time goes into drilling deeper.

Interview — 14 facets

Intent, scope, users, data, systems, constraints, don't-dos, risks, future, success metrics — scaled to the project's weight.

Readback gate

It restates the brief — goal, scope in and out, don't-dos, success criteria — and won't build until you say “yes, that's it.”

Scaffold the spec kit

constitution.md, spec.md, plan.md, tasks.md — written from your answers, wrapping GitHub Spec Kit for code projects.

Autopilot to the line

It builds as far as the stakes allow — QUICK builds all the way, FLAGSHIP stops at spec + plan for your sign-off.

Three weights. Autonomy scales inversely with stakes.

The more it matters, the more it stops and checks. The lighter it is, the more it just builds.

QUICKa hack / throwaway5 core facetsstops at: builds all the way

STANDARDa real project9 facets + readbackstops at: the task list — you say go

FLAGSHIPrevenue / public / clientall 14 + adversarialstops at: spec + plan — you sign off

WHAT YOU GET

A real brief, a real spec, and guardrails that bite.

A real interview

14 facets, scaled to the project's weight, with an explicit readback gate so nothing gets built until you confirm it.

A complete spec

constitution.md, spec.md, plan.md, tasks.md. Paste them into a repo, hand them to an agent, or keep them as source of truth.

A safety constitution

A tiered control system (Inviolable → Defaults → Preferences) that makes irreversible harm hard. Every rule carries its — Why:.

Enforcement that bites

On Claude Code: hooks + CI so “never commit a secret” and “never DROP TABLE without a backup” are gates, not hopes.

INSTALL IT

Two tracks. Pick one — or both.

The Claude Code track gives you auto-triggering and real enforcement. The chat track needs nothing but copy-paste.

Claude Code track — recommended

# 1. Clone the kit

$ git clone https://github.com/billy-mangaroa/interview-kit.git

# 2. Drop the skill into Claude Code

$ cp -r interview-kit/claude-code/skills/interview ~/.claude/skills/

# 3. In any project, run it

> /interview

Want auto-triggering on new projects, secret-scanning, and destructive-command guards? The full wiring (hooks, Spec Kit, branch protection) is in INSTALL.md.

Chat track — no install, any assistant

No setup needed. Open the prompt, copy it, paste it into a new conversation in Claude, ChatGPT, or Gemini, and answer. At the end it outputs your four spec documents as text.

Get the copy-paste prompt

USE IT EFFICIENTLY

Five ways to get the most out of it.

Name the weight honestly

Calling a real project QUICK skips the readback and lets it build the wrong thing fast. Match the weight to the stakes and let the depth follow.

Let it pre-load, then correct

It surfaces its assumptions before asking. Confirm or correct in one pass instead of re-explaining — that's where the saved time turns into deeper drilling.

Treat the readback as the moment

The restate-the-brief step is where you realise what you actually meant. Push back hard here — a mismatch now saves a rebuild later.

Brief another AI (or a contractor)

Run the chat track, get the four documents, and hand them over: “build this, follow the constitution.” No more lossy back-and-forth.

Adopt the constitution on its own

Even without an interview, the safety constitution + enforcement (secret-scanning, backup-before-migrate, destructive-command confirmation) prevents the most common solo-dev disasters.

Questions you probably have.

Is it really free?

Yes — free and MIT-licensed. Clone the repo, copy the skill in, and use it on a weekend hack or a client flagship. Share it freely.

Do I need Claude Code?

No. The Claude Code track gives you auto-triggering and mechanical enforcement, but the chat track is a single copy-paste prompt that works in Claude, ChatGPT, or Gemini with zero install.

What does it actually output?

Four documents: constitution.md, spec.md, plan.md, and tasks.md. On a code project it drives the real Spec Kit commands; on a non-code project it writes the same artifacts from templates.

Does it work for non-code projects?

Yes. It's context-universal — a video pipeline, a DJ-set workflow, an ops runbook. “User stories” become workflow steps and “acceptance criteria” become the done-signal.

What's the safety constitution?

A tiered set of non-negotiables (Inviolable → Defaults → Preferences) that the build runs under. Every rule carries its — Why:, so you can overrule it deliberately rather than blindly.

Interview yourselfbefore you build.