How I Use Claude/ChatGPT to Build a Real SaaS Solo

Most AI coding content stops when the demo works. That's not the hard part.

The hard part is getting from "looks cool" to live, safe, and able to take a real customer. I've done that across 4 shipped software products (+ 2 other product experiments). This is the process.

Exact prompts for each step → weekend build prompts.

1. Kill the idea first

Before code, I make AI argue against it:

• What's already out there?
• Why might this fail?
• Why would anyone switch?

Better to kill it in a chat than burn a week.

2. Build the ugly version

Usually Next.js + Vercel. Online fast. Not pretty. Not complete.

I only need to know:

• Does the data flow?
• Does the main path make sense?
• Can someone use it without me narrating?

Then I care about polish.

3. Let AI do the boring 80%

Auth, schemas, APIs, webhooks — not why I started the product.

I describe the outcome, let AI draft it, then review before it ships. Speed comes from clearing the boring layer so I can work on what's different.

4. Break it before users do

AI makes code *look* finished. That isn't finished.

Before real users or money: a hard security pass. Tell the model to attack auth, payments, cookies, data leaks. Insidr's pre-launch holes showed up here — full story.

5. Make billing boring

Upgrade, downgrade, cancel, refund — without emailing the founder. If billing needs me, the product isn't done.

6. Ship and watch what breaks

Weird production stuff only shows up live. Fix it as it appears. I've learned more from those moments than from predicting every failure in advance.

Prompts for every step: weekend-build-prompts.