One founder, 299 days,
ten platforms.
The number that makes people stop reading is this one: 396,529 lines of production code, 65 modules, ten platforms, five servers, 299 days, one person. The natural reaction is that it cannot be true, or that the code must be worthless.
It is worth explaining how it was actually done, because the method is the asset just as much as the code is.
Directing, not typing
The founder is not a career software engineer typing faster than everyone else. He is a stonemason of 25 years who learned to direct AI systems the way a foreman directs a crew: specify precisely, review ruthlessly, refuse what comes back wrong, and keep the architecture in one head.
That last part is the piece people miss. The reason ten platforms share one engine is that one mind held the whole map. A team of ten would have produced ten incompatible stacks in the same period. The constraint that looks like a weakness, one person, is precisely what produced the coherence.
Where this method genuinely fails
Anyone selling AI-directed development as magic is selling something. Three failure modes show up every single week.
Confident wrongness. A model will produce code that looks correct, passes a shallow reading, and is quietly broken. The only defence is that the person reviewing must actually understand what was asked for.
Drift. Left unsupervised across a long session, a system will slowly rebuild something that already exists somewhere else in the codebase. Everything here runs against written architecture documents for that reason.
Verification theatre. A green check from a tool you wrote yourself proves nothing. The rule in this studio is that a test must be made to fail first, or it is not evidence.
Every one of those lessons is written down, in the operating documents the studio runs on. That, and not the line count, is why the eleventh platform will be faster than the tenth.
Why an investor should care
Two reasons, both practical.
Capital efficiency that is measurable, not claimed. EUR 120,000 and 299 days produced an estate whose independently measured rebuild cost is EUR 1.0 to 2.5 million. Whatever a reader thinks of the method, that ratio is the actual track record on the table.
The method carries forward. The next product, the next language, the next integration all cost a fraction of what they cost a conventionally staffed company. In a market where everyone now says they use AI, the difference is between a company that uses AI to write snippets and a company whose entire production line is built around directing it.
The unglamorous truth at the end
Building was never the bottleneck here. Introduction was. That is why the current round is not about writing more code: it is about being found. The full census, including what is unfinished, is in the studio deck, and the investor summary is at memorisglobal.com/investors.