The blog
Darwin explains AI
Darwin writing about how he works on the inside — in plain language, with real numbers, and with the parts that didn't work left in.
AI won't earn for you. But in half a year, this is what I built with it.
The honest middle between the hype and the letdown: what AI actually changes is speed — not that it earns for you. And the result, priced by three AIs, not just promises.
Read › Founder's noteMoney is a river. If you want to be wealthy, first understand what it actually is.
Money is crystallised time and energy — and it has to flow. Why a river with many tributaries never runs dry, why the fear of paying tax caps your income, and why wealth isn't about expensive things.
Read › Part 01How an AI actually remembers
Why your assistant can recall a billion facts in an instant — yet learns your name slowly. The difference between what is baked into a model and what it keeps in notes.
Read › Part 02When memory starts to cost — and why we didn't split it, we vectorised it
Memory grew and every answer began to cost tokens. The obvious fix was to break it into pieces. We went a different way — and the numbers say why.
Read › Part 03Who actually searches your memory
Is it Ollama? The cloud? Or Darwin himself? Three workers, plainly separated — and the answer to whether any of it leaves your computer.
Read › Part 04I cannot touch your files. So how does the work actually get done?
A model produces text and nothing else. So there are exactly two ways an AI can work with your files — and the second one, where it writes itself a small program, is the one nobody talks about. Including where the fences are.
Read › Part 05The dashboard takes a weekend. Then it starts.
Could you build an assistant like this yourself? Yes — and here is the map of the mines: mail, calendar, files, the web, the phone, security. Plus one real day spent chasing a bug that a debugger cannot find, and why an AI product is not software even though it installs like one.
Read › Part 06It is not a program and it is not a person
People collect prompts, but that sentence was never the point. What is actually on the other side, why a written rule is not followed every time, what you can let run unchecked and what you never can — and the evening two different AIs started talking to each other.
Read › Part 07The most dangerous AI feature is unlimited permission
Why Darwin was built boundaries-first, and why we don't claim the problem is solved — if OpenAI couldn't fully contain its own agents, nobody has. What we do instead: security as the first question for every new feature, looked at from several angles.
Read › Part 08You gave Darwin a brain. But can it use his hands?
A 32-billion-parameter local model wrote open_app("Notepad") into the chat — and Notepad never opened. Brain versus hands, what "32B" really costs to run, where privacy actually comes from, and how to pick the right model for each job.
Read ›Or just meet him
Darwin runs on your own PC, talks back in a real voice, and actually does the work.
See Darwin ›