The Mullet Method
Why the smartest AI products don't look like AI at all
June 4, 2025
12,000+ automated tests. 95,000 lines of code. 53 days. Solo developer.
If you're building a product with AI, there's the easy way and the hard way.
The hard way is putting AI in front of your customers. You don't know what's coming in and you don't know what's coming out. Anything can happen, and it's all on your dime. Whenever your customers encounter a chatbot, LLM or some other thinking machine, you're doing it the hard way.
The easy way keeps customers away from AI entirely, through an interface that uses plain-old deterministic logic. Whether you click a button or scan a QR code, whatever happens next is predictable and testable.
Wait, where's the AI? Aren't you building a product with AI?
Yes -- with AI behind the scenes!
I use AI for:
Marketing Analysis: AI identified untapped opportunities to showcase QR-based experience cards within tourism, events, and personal branding markets. Database & Schema Design: AI designed a flexible database architecture that makes it simple to connect with enterprise systems, business data, and existing websites - unlocking valuable information previously trapped in hierarchical sites. Test-First Development at Scale: AI helped create thousands of tests throughout the development process, enabling a test-driven approach that evolved alongside the production code. Multi-Domain Deployment Automation: AI built deployment pipelines serving multiple business domains from a single codebase, handling environment-specific configurations across dev/test/prod. Performance & Monitoring: AI optimized QR token resolution to sub-500ms response times and created performance testing scripts that validate response times under load. Content & Template Generation: AI created domain-specific templates following industry design patterns and generated QR codes with precise positioning for print production.
Absolutely none of that involves putting a customer directly in front of a chatbot.
And because I'm building software with AI, the underlying logic can be highly nuanced. Every time you scan a QR code or press a button, you may get a different experience. That's not with AI in the loop, but rather the result of deliberate design choices about what should happen in different scenarios:
QR scans that trigger different content based on location, time, weather, and user history Button clicks that adapt to user behavior patterns without exposing any AI uncertainty Dynamic experiences where each domain gets customized templates and workflows
AI sets it up, and then the system works without the need for further intervention, just like an old-fashioned automaton in a mechanistic universe.
But if I want to change something, large or small, I have an evolved set of AI tools that can iterate, test, and deploy at dazzling speed.
I call it The Mullet Method -- Business Rules Up Front, AI Party in the Back.
And when you're rockin' a mullet, you're timeless.
Why This Works: The Hallucination Problem Last week I wrote about how AI "hallucinations" are really mismatched assumptions. When you ask AI to "pick a number between 1 and 10," it might give you π or √2 because you said "number" but meant "integer."
The same thing happens when building software. Ask AI to "write a program" and you'll get something that plausibly fits your request—but probably isn't what you wanted.
That's why The Mullet Method starts with tests, not code. Before AI writes a single line of production code, you ask it to write comprehensive tests that capture your exact requirements. No mismatched assumptions. No hallucinations. Just deterministic logic that does exactly what you specified.
The 11,909 test functions I mentioned? Those weren't written after the code—they were written before. I began each module asking AI to write tests that captured every edge case, every requirement, every assumption. Only then would I give it the okay to write code to pass those tests. (And whenever I took a shortcut, I soon regretted it!)
Think of it like the separation of powers in a constitutional democracy: The legislative branch writes code. The judicial branch ensures that all code passes preexisting constitutional tests along with other tests that evolve through experience. Then, you send your fully validated code to the server, which in this analogy is the executive branch. (Observe that having the server write and deploy untested code is an instant recipe for systemic failure.)
Same principle: List all the requirements (write the tests) before you try to build it (write the code) and definitely before you ship a product (deploy to server).
Result: A system that works exactly as intended, with no surprises.
I have a big announcement this week: I have 256 subscribers!
That's a cool million in binary!
But here's the thing -- if I can get to a million subscribers (a decimal million), I'm pretty sure I can monetize that somehow. And I need to crack the monetization code before my runway runs out.
Now, all modesty aside, I really think that I cracked the code on how to build software with AI.
You don't have to believe me. After all, I haven't shown you any working apps yet.
But here's what The Mullet Method delivered in just 53 days:
12,000+ automated tests across 640 test files 95,000 lines of code (64k Python, 34k Shell) 300,000 lines of documentation (3:1 docs-to-code ratio) 53 days from concept to working system
That's the kind of codebase you'd expect out of a small team of experienced engineers. But I built this whole thing myself. From scratch. Not bad for a solo dev.
Want to know how I did it?
Here's the deal: Every time this newsletter doubles its subscriber count, I'll drop a new chapter in my "How I Built It" series.
Get in early, and you'll be ahead of the curve. Join later, and you'll start from chapter one like everyone else.
At 512 subscribers, I'll release Chapter 1: "Mullet Method Foundations" along with an actual QR card that demonstrates the progressive unlock system.
At 1,024 subscribers, Chapter 2: "AI-Driven Architecture Design."
And so on, all the way to 1 million subscribers, where I'll release the complete source code with commercial licensing rights.
You want to learn how to build like this? Help me hit a million (decimal this time).
Let's rock the mullet together.
#AI #TechLeadership #SoftwareDevelopment #QRTechnology #ModelCitizenDeveloper