David Saa
Book a call →

The Ratchet — my proof-gated method for building software with AI

AI writes the code.
The gates decide if it ships.

AI will write you a thousand lines that compile, demo beautifully — and are quietly wrong. The Ratchet is how I make sure the ones that matter aren’t.

Try the wheel — it only turns one way.

Why you should care who’s driving

“Looks right” and “is right”
are different questions.

You’ve probably felt it, or you’re about to: AI now writes code faster than anyone can honestly check it — and some of it is subtly, expensively wrong in a way that looks completely finished.

It compiled. It demoed. Everyone nodded. And a week later you find the number on the screen was never real, or the thing that touches your money does something no one asked it to. The problem isn’t that AI is bad at code — it’s that speed hides the gap between looking right and being right. Most people answer that gap by hoping for a smarter model. I answer it differently.

I’m not new to method. I’ve spent 15+ years shipping digital products inside the disciplines software lives by — agile, design thinking, sprints and stage gates, from global brands to fintech. When AI changed the playing field, I did what you do when the field changes: wrote a new discipline for it. The Ratchet is that method — and it’s how one senior product person now designs, builds, and ships what used to take a team.

I didn’t learn this on a to-do app

I built software that handles real funds — and this is the method that made it safe.

The Ratchet came out of building a platform that manages liquidity-pool fee income and works to minimize impermanent loss for investors — software that handles real funds around the clock, the one kind where “looks right” costs actual dollars. When a mistake moves money, “the AI seemed confident” is not a plan. So I built one. Here’s what it produced, literally:

✓16 of 16 correctness tests passed before a single number was trusted on screen
✓every money-touching change attacked by a separate AI review before it could go live
✓every new capability ships switched OFF until a human turns it on

The method, in plain terms

Three promises I can actually keep.

Nothing ships on a model’s word.

Every stage of the work ends in written proof — a plan, a test table, a review verdict. If a number didn’t match a table, it doesn’t exist. No table means it’s a vibe, and vibes don’t ship.

Nothing goes live without your go.

Every new capability arrives switched off. At each checkpoint you get a plain-language briefing — what changed, what’s proven, what could go wrong, how it’s undone — and I take it live only when you say go. Small test first, always. An AI never makes that call.

Nothing depends on any one AI model.

The method assigns roles, not brand names — the skeptic, the builder, the workhorse. The AI models will change every few months; I’ve watched it happen mid-project. The process doesn’t flinch when they do.

“

Nothing I ship gets to overstate. A missing value shows a dash — never a made-up zero. When a result is bad, I tell you it’s bad. And when the honest answer to “should we do this?” is no, a clean no is a finished deliverable.

The rule the whole method is built on — never overstate, to you or in the work

Work built to that standard won’t flatter you with a number it can’t back. Neither will I. That’s worth more than any testimonial — because silent dishonesty is the thing that fails everywhere else.

Working together

What it looks like.

1

We scope the risk.

We find the parts of your project that actually matter — the ones that touch money, customers, or production — and treat them differently from the rest.

2

I build gated.

You watch the proof accumulate — plans, test tables, review verdicts — as each stage clicks into place. No black box; you see the receipts.

3

You say go.

When the proof says it’s safe, you give the go and I take it live — small test first. You’re never asked to trust; you’re shown why you can.

This isn’t a process I bolt on for show — it’s how I work on every build, the discipline behind everything I ship for you. And because it leans on roles instead of brand-name models, it holds up as the AI landscape keeps changing.

Start the Ratchet — Stage 1: Scope

Your build’s first click,
right now.

Five quick questions — the same ones that open every real engagement. No email needed to begin, and your answers stay in your browser until you decide to send them.

Each answer clicks the wheel forward. It doesn’t turn back.

QUESTION 1 / 5

What are you building?

Nothing is captured until you hit send. A dash, never a fake zero — that applies here too.

Let’s talk

If you’re about to trust AI with something you can’t afford to get wrong —

talk to me first. Bring me the project that has to be right, and I’ll show you how the proof accumulates.

Book a 30-min call →
Prefer email? ux@davidsaa.com · See ways to work together · or start with Stage 1 above