A team taught AI to read 1,699 legacy classes and turn them into living models
AI read 12 modules of a municipal ERP, captured its business rules as formal models, and regenerated 3.1 million lines of working code.
This team looked at a 20-year-old municipal system and saw not a burden but a store of knowledge waiting to be freed. They taught AI to read real production code, pull out the rules hidden inside, and rebuild whole applications from a single model. One of them did it at 62, which says a lot about when it is too late to build: never.
The Modeler Studio — Teaching AI to Read Legacy Code and Preserve Business Knowledge
Before
They worked with a municipal ERP whose business knowledge was buried deep in production source code, spread across 12 modules and 1,699 classes. The rules that mattered, the formulas and derivations, were locked inside code that had grown over two decades. Old systems like this are usually treated as a liability to be replaced, not understood.
“A 20-year-old system is not a liability; it is a knowledge asset waiting to be freed.”
In their own words
The turning point
They set out to teach AI, using Claude, to do more than skim code. The key shift was giving it a rigorous formal destination: a four-step method of Read, Learn, Capture and Regenerate. Instead of asking AI for vague understanding, they asked it to write business rules into an industry-standard conceptual model, where the model is the system, not documentation.
“Structure beats prompts: giving AI a rigorous formal destination turns vague "understanding" into verifiable knowledge.”
In their own words
After
AI analyzed 1,699 classes and found 310 containing mathematical business rules, captured in about 18,500 lines of documented formulas indexed in a vector database. They built their own generation engine in C#/.NET, more than 70,000 lines, that turns that model into working applications. So far it has produced 3.1 million lines of code and 7 live applications across Blazor web, REST API, Windows desktop and React.
What it lets people do
Organisations with aging software can recover the business knowledge inside it and keep it as a lasting asset. From one model, working applications can be regenerated in whatever technology comes next, so the rules survive even when the code is replaced.
What they learned
They came away convinced that code is disposable and knowledge is what lasts, and that real business intelligence lives in valuations and derivations rather than transactions. Above all, structure beats prompts: give AI a precise place to put what it finds, and understanding becomes something you can verify.
“It is never too late to build — one of us did this at 62.”
In their own words
The Modeler Studio — Teaching AI to Read Legacy Code and Preserve Business Knowledge was built during AI Mastery.
Read the full submission
Written 7 August 2026 Submitted by the storyteller
Every story is written by a Mindvalley student or by someone our teachers have worked with. How we collect and check stories
More from AI Mastery
