DEV ENVIRONMENT — TEST PAYMENTS ONLY · agent-edited · not production
MFounders Crash Course · 45 min
Founders Accelerated · Tuesday 2026-04-28

Years of no-code AI
condensed in 45 min.

Four modules. The smallest set of moves that lets you ship a real product without writing the code yourself.

Websites + GitHub Supabase backend AI coding agents Hosting + OpenRouter
Why this set

You don't write code.
The agent does.

The wrong path

  • "I'll learn to code first, then build."
  • "I'll pay for SaaS forever."
  • "I need to hire a developer."
  • Months of friction, no shipped thing.

The path that works

  • Learn the 4 moves below well enough to direct an agent.
  • Agent reads your docs, edits files, runs commands.
  • You review + decide what ships.
  • One weekend = a real product live on the internet.
Module 1 · 8 min

Websites + GitHub.

Every change is a snapshot. Every mistake is reversible.

Git: snapshots, not diffs.

Commit

A snapshot of every file at this exact moment. Always recoverable.

Branch

A parallel timeline. Try something risky on a branch; merge if it works, abandon if it doesn't.

Rollback

"Take me back to yesterday's snapshot." One command.

GitHub = cloud backup + collaboration on top of git.

The three commands you'll use.

# save your work as a snapshot git add . git commit -m "updated homepage copy" git push # oh no — the change broke something git revert HEAD # undo the last commit cleanly

The agent runs these. You decide when to commit + when to roll back.

Module 2 · 12 min

Supabase
= your backend.

Database + auth + storage + functions. One platform, no servers to run.

Four things, one dashboard.

Database

Real PostgreSQL. Tables, queries, indexes. Auto-generated REST + GraphQL APIs from the schema.

Auth

Email/password, magic link, Google, GitHub. JWT-based sessions. Row Level Security ties data access to user identity.

Storage

S3-style file storage with the same RLS-aware access rules as the database.

Functions

Serverless TypeScript functions for anything the database can't express directly. Run on demand.

The thing that bites everyone

Row Level Security.

The default

A new table is readable by anyone. Your members' emails are publicly queryable until you enable RLS.

The fix

For every table: ALTER TABLE ... ENABLE ROW LEVEL SECURITY; + write policies that say "members read their own row, founders read all".

Every MAH table has RLS. Confirm it on your own projects too.

Module 3 · 15 min · the centre of gravity

AI coding agents.

They edit files. They run commands. They commit + push. You direct.

The three agents we use.

Claude Code

Interactive terminal agent. You sit with it; describe what you want; it edits files, runs tests, commits. Your daily driver.

Hermes

Autonomous Mac daemon. Reads Telegram, runs scheduled jobs, drafts content overnight. The compounding background worker.

OpenClaw

The pattern: a Federation of agents collaborating. Each one has a role (operator, coder, researcher). Scales beyond one human.

The mindset shift

You're the operator.

Old mental model

  • I have to learn to code first.
  • I type the syntax myself.
  • I'm the bottleneck.
  • One project at a time.

New mental model

  • I describe what I want, in plain language.
  • The agent types the syntax.
  • I review + decide what ships.
  • Multiple agents running in parallel.

How to direct an agent.

# 1. Context — what's the situation? The MAH members area uses Supabase. The Learning tab loads mah_assessments by email. We just added a co-founder branch. # 2. Constraint — what must be true? Don't break M40H (DM module). Cache-bust new assets ?v=m41. # 3. Success — what does done look like? A new card in panel-learning showing the user's latest score. # 4. Verb — go. Write js/members-courses.js. Wire it. Test it.
Module 4 · 7 min

Hosting + OpenRouter.

Get the site live. Pay less for AI calls. Both via the agent.

Pick a host, plug into the agent.

Cloudflare Pages

Free for static sites. Connect a GitHub repo, every commit auto-deploys. Use for marketing pages and docs.

Hostinger

Cheap shared hosting for small dynamic sites with PHP. ~$2-5/month. Has API access.

AWS Lightsail

Where MAH + TCS live. ~$5-20/month per server. Full control. Talks to the agent via SSH.

All three can be driven by the agent — you describe the deploy, it runs the commands.

The single biggest cost-saver

One key. Every model.

Why use it

One API key, one balance, route to the cheapest model that's good enough for each task. Spend $20/mo instead of $200.

What you get

Claude (Anthropic). GPT (OpenAI). Gemini (Google). Open-source models on Together / Fireworks. Free tier on some.

How it plugs in

Drop the OpenRouter API key into your project's .env. Most agent tools support OpenRouter natively.

How it all fits.

You
describe
Claude Code
edits files
git commit
+ push
Hosting
auto-deploys
Supabase
stores data
OpenRouter
powers AI calls

A real product, end to end, with you directing every step but typing none of it.

By the end of this

You can run the build → ship → revert loop solo.

You'll be able to

  • Open a project, describe a change, drive an agent to make it.
  • Review the diff (just enough to know it's not crazy).
  • Commit, push, watch the deploy.
  • Roll back in one command if it broke.

You won't need to

  • Write the code yourself.
  • Memorise SQL or HTTP details.
  • Wait for "the developer."
  • Pay forever for SaaS that locks your data.
Tuesday 2026-04-28 · 6:15 PM AEST · 45 min

What we'll actually do Tuesday.

00:00 - 00:10

Quiz delta walkthrough. Where you scored, which modules need depth.

00:10 - 00:30

Live demo of the build-ship-revert loop on a real MAH change.

00:30 - 00:45

Logins for GitHub, Supabase, Claude Code, OpenRouter. Each cofounder leaves access-ready.

Take the readiness check before Tuesday 5pm.

melbourneaihub.com.au/founders-readiness.html