From 36% to 100%: How Self-Improving Agents Write Their Own Skills — Rafal Wilinski, Runlayer
Agents that write and share skills can lift task success from 36% to 100% enterprise-wide.
“a skill is an instruction that an agent can read when they feel that knowledge is relevant”
Rafal Wilinski of Runlayer presents a pattern where agents automatically write reusable 'skills' — progressively-disclosed instructions — when solving hard problems, so future agents inherit that knowledge rather than starting from scratch. The core insight is that most agent knowledge today evaporates at session end, and structured skill persistence can compound improvements across an entire organization. The 36%→100% headline frames this as a measurable reliability upgrade, not just a theoretical architecture.