Free · v1.1 · For non-technical founders using AI

Ship real software with AI. Without being an engineer.

You've been building with Claude, Cursor, or Lovable. Things kind of work, then they break. You don't know what questions to ask. You feel like you're one step away from a real product but can't quite get there.

Action Building is the playbook that gets you over that line. From a founder who built six real products in six weeks using the exact same tools you're using.

Get the pack → What's inside

If any of this sounds familiar, you're in the right place.

Everyone says you can build software with AI now. Nobody tells you what to do when it stops working. We did the thing nobody has done yet: we wrote it down.

None of this is your fault. AI tools are amazing at demos and terrible at production. The missing piece isn't more code. It's the practice of building. Action Building is that practice.

How one founder ships like a team

The answer isn't more code. It's three things, installed on Day 0.

A 200-person engineering team exists because no single engineer can hold auth, schema, deploys, migrations, reviews, incident response, and security in their head at once. A single founder with the right pack can hold all of it, because the pack externalizes the memory into artifacts the AI pair reads. Six products, six weeks, one founder. Here's how.

01 / Buy the plumbing

Don't rebuild what's already solved.

Auth, databases, hosting, payments, email, moderation. Clerk, Supabase, Vercel, Stripe, Resend cover the infrastructure that used to require engineering teams. You integrate. You never rebuild. One founder integrates; zero founders rebuild.

02 / Install discipline Day 0

Not Day 90. Not after the first incident.

Darby took 44 days because the practice was discovered through incidents. DeepKimchi took 8 days because the same practice was installed on Day 0. The governance-latency curve is measurable and monotonic. The pack is that practice, pre-installed.

03 / Stay conversational

No CLI. No Docker. No local dev environment.

MCP connectors replace the terminal. Every infrastructure operation runs from inside the AI conversation. Your conversation history becomes your audit trail. Your AI pair chains operations without context-switching. You stay the founder.

In the ZIP 37 skills · 37 pattern pages · 15 governance rules · 8 templates · 4 case-study journeys · a single 335 KB file

The proof

Six real products. Six weeks. One founder with an AI pair.

Every pattern in the pack came from one of these six projects. These aren't demos or prototypes. They're running in production, used by real people. You can verify every one.

Darby

A healthcare AI that reads prescriptions from faxes and turns them into orders. Built for medical equipment suppliers who drown in paperwork.

Live today· Built in about six weeks

twi5t

A sales-automation tool that finds prospects, drafts outreach, and tracks responses. Robin uses it every day to grow his own companies.

Live today· Built in about six weeks

blocktim3

An AI image generator for storybook illustrations. On its busiest day it shipped 22 real updates to users without a single rollback.

Live today· Built in under a week

DeepKimchi

A dating app for people who want to date across cultures, with an AI wingman that helps you message better. iOS and Android.

Coming to the App Store and Google Play· Built in 8 days

celltracker

A sample-tracking tool for a biotech lab. Small, practical, real. Lab uses it every day.

In use inside Wonderlab Bio· Built in under 2 days

Wonderlab Bio

The holding-company site. Nothing fancy. Just a clean, well-organized page for the portfolio of companies.

Live today· Shipped and maintained over about three weeks
RYS

Who made this

Not a software engineer. A founder who kept hitting the same walls you're hitting.

Robin Y Smith is a chemist-turned-serial-founder with nearly 30 years of building software and life-science companies. He invented the electronic lab notebook and holds 23 issued US patents. He founded and sold Synthematix (acquired by Symyx, 2005) and ArtusLabs (acquired by PerkinElmer, 2011), the two companies that pioneered the digital lab notebook category. He later founded Orig3n/Seaport Diagnostics (acquired by KKR, 2021). Today he runs Wonderlab Bio and Darby, an AI-first HealthTech company, from Boston, where he's using Action Building to build and run several companies in parallel. EY Entrepreneur of the Year New England finalist, 2018.

I built Action Building because I kept hitting the same walls as every non-engineer trying to ship with AI. The pack is the version of this practice I wish had existed when I started.

LinkedIn X / @robinysmith Wonderlab Bio Darby

How it actually goes

One afternoon. Here's the whole thing.

  1. Download the pack.

    One ZIP, 335 KB. Free. Just your email.

  2. Drag the ZIP into Claude Code.

    (Or Cowork, Cursor, Windsurf. Any AI pair that reads files.)

  3. Claude reads it and asks you three questions.

    What's your project called. What does it do in one sentence. What tech stack, or should Claude recommend one. You answer in normal words.

  4. Claude walks you through Day 0.

    Your architecture plan, your managed services (Clerk for auth, Supabase for database, Vercel for hosting), your database design, your deploy pipeline. Claude does the work. You review and approve.

  5. Claude hands you a ZIP.

    You drop it into a GitHub folder. Commit in GitHub Desktop (the app, not the terminal). Push. Vercel auto-builds.

  6. By dinner, your site is live.

    Real database, real auth, real deploy pipeline, real rollback button if something breaks. The foundation of a real product.

Every day after this uses the same loop. You talk to Claude. Claude produces a ZIP. You commit. Vercel deploys. That's Action Building.

§ Why we call it Action Building

The canvas was not a surface. It was an arena.

In 1952, the critic Harold Rosenberg coined the term Action Painting to describe what Jackson Pollock and his contemporaries were doing. The canvas, Rosenberg argued, was not a surface to represent something. The canvas was an arena in which to act. The painting was not a product. It was the record of an event.

Critics dismissed Pollock for decades. Anyone could throw paint, they said. Then physicists analyzed his drip paintings and discovered fractal mathematics embedded in the patterns: structures that could not be replicated by random splashing. What looked chaotic was deeply, measurably intentional.

Action Building. That is exactly what we do.

The conversation is not a planning tool. The conversation is the arena in which the software is built. The codebase is not separate from the dialogue. It is the record of the dialogue.

6

products shipped

6

weeks

1

founder

1

AI pair

People will say "anyone can talk to an AI." They said the same thing about Pollock. Until they tried to replicate the fractals.

Robin Y Smith · Action Building · 2026

Get the pack

Sign up to download the pack. Free.

I'm building Action Building as free education for founders like you: people who are trying to ship real software with AI and running into the same walls I ran into. Your email does two things. It gets you the download right now. And it lets me tell you when v1.2, the book, and new pattern drops go live. No spam. No upsell. Your address stays with me and nowhere else.

action-building-cold-start-v1.1.zip

The Cold Start Pack.

37 skills. 37 pattern pages. 8 templates. Four case-study journey docs. A full Day-0 playbook. 335 KB.

v1.1 · 2026-04-19 · Works with Claude Code, Cowork, Cursor, Windsurf.

Before you sign up

Is this for you?

This is probably for you if…

  • You're a non-technical founder. Product manager, designer, domain expert, operator, founder who isn't an engineer by trade.
  • You've been using Claude Code, Cursor, or a similar AI pair to build something real.
  • You've hit walls that feel like "I should know this but I don't." Databases, auth, deploys, secrets, scaling.
  • You want to ship a real business, not a side project. With paying users, real data, real stakes.
  • You're smart and you're willing to learn. You just want to learn the right things, not code-101.

This is probably not for you if…

  • You've never built anything with AI. Go try Lovable or v0 first. Come back when you've hit the walls.
  • You're an experienced software engineer. Some of this will be useful. Most will be obvious. Skim the patterns; skip the basics.
  • You're looking for a coding tutorial. This is a methodology, not a course. You'll still need Claude to write the code.
  • Your product doesn't need persistence. If you're building a static site or a one-off tool, you don't need most of this.

Questions people ask…

  • Is it really free? Yes. No freemium trick. MIT license. Your email gets you the download and the occasional update. That's it.
  • Why do you need my email? Two reasons. One, we send the link to your inbox so you have it for later. Two, when v1.2 ships with new skills and patterns, you'll be the first to know.
  • Will it work with Cursor or Cowork instead of Claude Code? Yes. The skills and patterns are plain markdown. Any AI pair that reads files can use them.
  • Is there a paid version coming? Eventually, a book and a hosted analyzer. The pack itself stays free.
  • What if I get stuck? Honest answer: we're early. DM Robin on X. Contributions welcome.

Start your Day 0.

Grab the pack. Open Claude Code. You'll be shipping by this evening.

Get the Cold Start Pack →