An Absurd Stone

An Absurd Stone has no goal to reach and nothing to complete. You simply spend time with a stone through ordinary days, changing moods, and passing rain, and one day it may leave. With no task to optimise, I shaped its features through game design instead, thinking about pacing, rarity, surprise, and attachment.

I defined the core loop: meet a stone, touch it, watch it respond and slowly change, then return without a task to complete. Claude Code and Codex helped assess feasibility and turn that interaction into a native MVP. They proposed implementation paths; I chose what to pursue and judged each result against the product's intent.

Figma established the interface and interaction foundation. I translated those decisions into Markdown covering the design system, product behavior, voice, and current project context, so Claude Code, Codex, and collaborators could work from the same language. My friend Shi Qi, a 3D designer, also wrote code with me, and files like AGENTS.md and progress.md kept both of us, and every agent, on the same context, so either of us could pick up where the other left off.

Shi Qi and I explored the possibilities, trying ideas and defining how the stone should respond, from how snow settles and rain darkens the surface to how small animals move across it. She then went deeper into the models themselves, solving technical questions around modelling, anchor points, materials, and textures, and set the base rules for the first three stones. Much of the automation that followed was built on that foundation.

When a task began to repeat, from shaping the stone and preparing 3D models to reviewing code, performance, security, privacy, and design consistency, I wrote it down as a Markdown file, a Skill or documented process that an agent could follow on its own. Each one turned a step we used to handle by hand into one that simply ran, and that automation was the key to how fast the project moved.