7 pages
Karpathy, Tan, and Liu all start from the same diagnosis — your agent is a retriever, not a thinker. They reach three different architectures: retrieve (RAG), compile (LLM Wiki), and act (Fat Skills / GBrain). Trail picked Compile, deliberately.
Anyone can claim an AI-generated feature works. The interesting question is whether anyone has actually watched it do that — in a real browser, not just a log line that says "build succeeded."
We never route an AI task to a specific model because it's the one we're used to. We route by the task — and the moment it touches your customers' data, it's automatically sent to a model hosted in the EU.
broberg.ai isn't built by one developer at a time. It's built by a fleet of AI agents working in parallel, checking each other's work, and never calling anything done until it's been proven in a real browser.
The engine that drives every project in the universe — from first idea to finished, live product.
Every AI call in the broberg.ai universe goes through ONE facade — Claude, GPT, Gemini, Mistral, Black Forest Labs.
30 years of experience building software — and a healthy scepticism that means I do not shoehorn AI in where it does not belong.