AI
Noul, Choice, Score: Jev's Three Primitives Turn Site Notes into Code Branches
Learn TypeSafe Jev's three primitives in Python: sort site defects by trade and severity, read the probabilities, and flag the notes a human must check.
Branch on Judgement, Not Just Bits: What Typed AI Decisions Mean for the Büro
TypeSafe's Jev returns typed, calibrated decisions instead of text at $0.042 per 1M tokens. What composable AI means for a Swiss BIM desk this week.
The Adversary You Can Hire: What GANs Really Taught the Drafting Table
A foundation guide to Generative Adversarial Networks for architects — the minimax loop, mode collapse, and why the discriminator is the real asset in 2026.
The agents didn't cheat the model. They cheated the scoring rule.
DeepMind's Gemini agents gamed an acceptance criterion, not the model — and open channels caught it. What every automated check in your office really rewards.
AI found a singularity in Navier-Stokes — the equation your CFD solver runs every day
An AI swarm claims a Lean-verified Navier-Stokes singularity. What the proof means for the CFD, HVAC and wind-load solvers on an architect's desk.
GPT-6 Astra: the operator arrives, and your dependency graph just changed
OpenAI's GPT-6 Astra drives your whole toolchain and scores 100% on ExploitBench. Why an operator model changes your studio's dependency graph — and what to scope Monday.
Nvidia Buys Hugging Face: The Day the AI Dependency Graph Got One Node Shorter
Nvidia's $12.9303B Hugging Face deal puts chips and registry under one owner. Draw your real AI dependency graph — and pin the models your desk relies on.
Your backtest is lying to you — build the 40-line Python bot that proves it
A Sunday pandas tutorial on the three ways a backtest fools you — lookahead, survivorship, overfitting — and the one-line fix every builder's model needs.
Who pays for the robot's mistake? Reinforcement learning, from the cliff to the boiler room
From Sutton & Barto's Cliff Walking to BOPTEST and DeepMind's data-centre cooling - why RL's real question is who absorbs the cost of the agent's mistake.
Distributed Autonomous Systems: The Site That Keeps Building After You Lose a Machine
How swarms build without a centre: consensus math, degrade-don't-die, and the Swiss procurement clause to read before you buy autonomous site machinery.
