Martin
.
Blog
Neurons
Workshop
Podcast
About
Contact
Meet Martin
DE
/
EN
Newsletter
← All articles
Net
Timeline
Reset
From the Workshop: Did this model learn a game? An attempt from the workshop: optimising models without training them.
From the Workshop: At night the platform deliberates with itself. Two workshop days, seven findings, and a question we cannot answer.
From the Workshop: MCP, second attempt: 48 tools, three channels. What we learned in October.
AI · Opinion: Simply put: frontier AI. Why the most intelligent AI is not always the best solution.
AI · Opinion: Follow-up on the Anthropic prospectus: $895 billion in commitments, $3.2 trillion at the big five. Which of it is debt, and which is not.
Security: The model wasn't hacked. The environment we hand our development rights to was.
Security: A valid signature is not yet a safe decision. What I learn from the Binance incident.
AI · Opinion: The Anthropic prospectus: AI is getting more efficient. But who knows the real costs?
Customers: Enterprise AI: We do not need more chatbots. We need to finally solve problems.
AI Driven Development: "Build me a great contact manager." What the AI then builds, and why that is not the system.
From the workshop: A day in the workshop. Nine bugs, one typed sentence, four training signals.
AI · Opinion: Reverse engineering for everyone. When copying costs nothing, what is your code still worth?
From the workshop: 169 files. The day my agent wanted to deliver too much.
AI · Opinion: The most dangerous AI may be the one that wants to help us.
AI · Opinion: Agents drift before they break out. What comes to mind when you shape models and agents.
AI Driven Development: AI writes 30 percent more code. And who said more code is better software?
AI · Opinion: Who owns the code the machine writes? Not the provider. But perhaps not you either.
AI · Opinion: Volume and speed. And who still builds something that runs in ten years?
Thoughts: Flip book or current chart. Why the time between two commits is the real signal.
Read for you: Read for you: "Harness Engineering". Old rules, new name, one good point.
From the workshop: I tried to train myself. It works. Partly. Sometimes well.
AI Driven Development: Software is not a photo. It is a film. And no tool watches it.
AI Driven Development: We wanted to develop faster. Now we only press Enter.
AI · Opinion: MCP is slow. That is not a bug, that is the blueprint. What I learned with 33 tools.
Workshop: One photo, one face that speaks. How we built a mini model for an avatar in the workshop.
AI · Opinion: Faster than the rules. The question I keep asking myself since the models started helping with their own training.
AI · Opinion: Can an AI really escape and make itself better?
From the workshop: Test data without data: how to compare production against test without anyone seeing the customer data
Software · Agents: Code is getting cheap. Knowledge isn’t.
Agents: “We want an AI agent”. The wrong requirement
Projects: Our first model, 2020: object recognition in the living room. And why Xcode 2026 hangs on it.
AI Driven Development: Innovation workshop: The three questions that immediately weed out 80 percent of AI ideas
AI Driven Development: From 20 developers to 5: How to rebuild a team without losing it
AI Driven Development: The first conversation with a CTO who wants to "just introduce AI"
AI Driven Development: I don't just build AI. I take teams to the place where they can work with it.
AI Driven Development: The architect is not dying out. He is becoming more important than ever.
AI · Opinion: Nvidia is almost giving GPUs away. And why that won't last
Personal: Built, not posted: why I am writing after all
Agents: Contract first: why our backend and frontend agents aren't allowed to look at each other's code
Agents: An agent must never sign off its own work. The one rule that saved NeoCoder
From the workshop: Keeping 50 micro-models alive: why automation isn't optional, it's mandatory
Projects: Smart Vending: What a vending machine teaches about AI at the edge of the network
AI · Opinion: Factor 400,000: why anyone who only pays for tokens doesn't know the price of AI
Projects: YOW: one year, three people, one catapult. And every angle, every weight really computed.
Projects: Dunk Museum for Titolo: a storyboard, Unity in the browser, and a budget nobody understood
From the workshop: 80 MB for Apple: What shrinking a model taught me about AI
Projects: Scan Your World: the attempt after. LiDAR, our own tracking, and a why I cannot fully piece together anymore.
Projects: SKIMO: the fridge that knew what was inside. And where we went from a small model to a bigger one.