• The flat space problem: planarization in vector retrieval

  How we discovered that our embedding space had collapsed into something nearly one-dimensional — every query returning the same confident cluster — and the three things we did to fix it: a genetic algorithm on the tag vocabulary, a pastoral index, and a council of per-author agents.

• Slow on purpose: private audio analytics without real-time

  Five constraints for privacy-preserving ambient audio — cheap hardware, no stored audio, no identity, a fuzzy signal, and no deadline — look adversarial and turn out to reinforce each other. Relaxing the deadline is what pays for the rest.

• The bowl only holds what it received: critical chain scheduling

  Critical chain project management prescribes rules that look arbitrary — one pooled buffer, a ban on multitasking, no fixed formula for the schedule. Five results from probability, queueing theory, and computational complexity say otherwise.

• Direct manipulation and the shape of AI interfaces

  Directness — putting the object of work on the screen and acting on it — was an under-acknowledged engine of the graphical interface, and conversational AI runs against it. A rubric to measure it, the strongest objections tested fairly, and a footbridge designed inside a chat by a real finite-element solver.

• The design language of AI: from terminal to cinema

  Because AI arrives as text, we dress it as a terminal — but a terminal is a console, not an intelligence. The real design language of AI is the union of text and photorealism in a single frame, a fusion cinema imagined long before anything could render it. Grounded in a corpus of 565 design references.

• Compatibility of worlds: encoding facts as binary vectors

  An eight-year-old cognitive-science question about crossover fiction, traced through Leibniz, Carnap and Kripke, and revisited with binary vectors borrowed from hyperdimensional computing.