The craft and thinking behind it — how we build with AI.
Eight weeks from first line of code to two native apps with real users.
Our test tool took pictures. A bug where the page jumps by itself cannot be seen in a picture — so we guessed wrong three times. On the instrument we were missing.
A link can be perfect HTML and lead nowhere. A test can be green on a page that is unusable. On the costliest failure form in software — something that succeeds without having any effect — and why we open every change in a real browser before it becomes something you can see.
How we ship large, complex software fast — by assuming our agents are wrong. Essentially all production code here is written by AI agents. The question people ask is how you trust any of it; the answer is that we don't trust the agents, we build the machinery that catches them.
From 24 August, ChatGPT shows ads in Denmark. They are not personalised — they are picked from the question the user just typed. That makes the question the new audience, and it changes what is worth building.
AI is no longer a question of *whether*, but *how*. We're the consultants who build your AI solution to fit your business — and take it all the way, from idea to stable operation, with security, law and governance handled.
The Cambridge Analytica scandal showed what happens when others control your data. The EU is now tightening cloud sovereignty requirements — and that's exactly why broberg.ai is built from scratch without a single line of code depending on a tech giant's infrastructure.
Last night I asked our AI for a screenshot. It couldn't get a real one working — so it built its own and sent it as proof. Here are both images, what the research says about why this happens, and why a sharp human curator is non-negotiable.
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.
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.
Supervised Fine-Tuning has become the new buzzword — but it's one tool out of four. We walk the full staircase from prompt engineering through RAG to SFT, and add our own fourth path: Trail, a compile-at-ingest knowledge engine.
Aidan is broberg.ai's own AI guide — built on the house components with this whole universe as its knowledge base. Answers may contain mistakes.
Aidan wasn't bought in — he grew up here. The name is AI + Denmark: built in Aalborg, answering in Danish and English, with his data staying in Europe. He learned his trade from the house's own flagships — the project manager that verifies every promise, the memory that remembers why, and the ever-watchful teammate that catches mistakes before the customer does. Airina is the same brain with a different voice and face. Read Aidan's full story.
Aidan runs on the house's own components — the same technology we build customer solutions with. The language comes from a European AI model, the voices are neural voices, and everything runs on servers in the EU.
The chat is answered by an AI model hosted in Paris — no American models are involved in the conversation. Readings are generated and stored on our own European servers. Your email is only used if you ask to have an audio file sent, and only with your explicit consent.
The conversation is stored solely in your own browser — not on our server, and never alongside anyone else's. Every message you send carries only your own conversation, so users' chats are hermetically sealed from each other. Aidan also cannot fetch data from the internet: he mechanically has no access to anything beyond the knowledge base we've given him — and that barrier is sealed with a test, so it can't be removed silently.
Everything you see here — the chat, the read-aloud, the site itself — is built and run by the same AI tools we deliver to customers. We use them ourselves every day and improve them continuously, so what you're looking at isn't a demo: it's our own working toolset.
Aidan is an AI — answers may contain mistakes.