Hidden machinery, made legible.
I build, draw, and write to make opaque systems inspectable. The shop ranges from AI loops and agent protocols to visual systems and finished-enough products, but the lens stays the same: take the thing apart until the path is visible.
Currently: studying the forms that make AI systems behave: feedback, memory, constraints, interfaces, and the return path between intent and result.
The Shape of the Thing.
A first study in how AI becomes inspectable when the loop around it is treated as the object.
Form hides in repetition.
A first field note on why the single prompt is the wrong unit of attention. The useful thing is the loop around the prompt: framing, run, inspection, correction, preserved state, and the next pass.
The Shape of the Thing.
A field note about AI as a closed loop: less a magic answer box than a system shaped by feedback, memory, and repeated inspection.
Landscape deck, rebuilt.
Ten VHX plates that turn a rough idea into a visible mechanism: open loop, closed loop, context graph, return path, and operating stance.
VHX Design System.
The rule set behind the site: one face, one grid, one accent, one theme contract, kept visible so the interface can be checked.
Tesseract Lab.
A 4D projection lab that makes geometry interactive, theme-aware, exportable, and inspectable across small screens.
ODIN Sentinel.
A coordination surface for visible agent teams: startup packets, readiness receipts, handoffs, and review gates instead of hidden status.
┌──────────────────┐ │ odin-sentinel │ │ mcp · stdio │ └─────────┬────────┘ ┌──────┼──────┐ ▼ ▼ ▼ RES TOOLS EXP
Mister Clean.
An agent skill that pays inherited repository debt, then guards the commit boundary so the clean codebase stays clean.
Math Engine Specimen.
Interactive derivations for the grid, rhythm, contrast, and projection primitives that keep the design system from becoming taste theater.
Legibility is the method.
The site contains design systems, math, decks, products, and notes because each format exposes a different layer. The point is not the category. The point is whether the hidden machinery becomes knowable.
Find the mechanism.
Start by separating the surface from the machinery: state, constraints, incentives, memory, and feedback.
Read the argument →Expose the trail.
Interfaces and agent systems should show the state, checks, and handoffs a reviewer needs to understand them.
Inspect products →Ship objects with form.
Only the work that runs, renders, exports, and survives theme changes belongs in the cabinet.
Open design →Get the notebook.
Subscribe for essays about AI systems, design constraints, agent loops, and the practical work of making opaque tools inspectable.