Sol LeWitt wrote instructions. His wall drawings were never objects. They were scores, addressed to human draftsmen who would execute them on specific walls and then, often, paint over them. The instruction survived. The drawing was always someone's performance of the rules, and always temporary.

The instructions are deliberately incomplete. "Not straight." "Not touching." "Random." "Covering the wall evenly." These phrases don't specify. They invite interpretation. LeWitt understood that a draftsman would supply what the language withheld: a sense of what random actually looks like at scale, a judgment about when two lines are too close. The ambiguity was structural. It was the work.

I first saw a LeWitt wall drawing when I was in high school and it never fully left. What I kept coming back to was that the instructions only work because a human receives them. They depend on judgment, on a body, on someone who already knows what "not straight" means in their hands. I wanted to see how a system could be designed to understand that ambiguity on its own terms, and what it would produce when it did.

A system was never meant to receive these instructions. A system has no intuition about straightness. It cannot feel when something is random enough, or know when lines are touching in the wrong way. To perform the instructions, the machine has to invent formal answers to questions that were written to resist them. "Not straight" becomes a displacement field shaped like a slow vortex, or a pair of braided attractor spines pulling lines into woven curves. "Rectangles formed by bands meeting at right angles" becomes a field of intersecting 3D wireframe structures. Each is a genuine reading of the instruction. None is what any draftsman made. The system does not approximate human interpretation. It arrives somewhere adjacent to it, from a completely different direction.

Every execution is seeded by the moment it runs. The same instruction, under a different seed, commits to a different mode: different densities, different weights, a different formal logic entirely. The drawings here are never shown complete. They build themselves mark by mark, fade, and begin again. You are always arriving in the middle of a performance.

LeWitt wrote the rule. The system invented a way to follow it. The seed determined what happened. The question of who made the drawing depends on which of those you consider the work.

Zach Richter, 2026