Diffusion of Innovation: How Agentic Skills Spread

Session track: AI Engineering

Session time: 11:30 pm - 12:00 am UTC November 13

Session description:

When one AI agent discovers a way to fix a broken build, should others adopt it? What if the shared procedure helps one agent but worsens another’s results? Inspired by Everett Rogers’s Diffusion of Innovations, this session follows agents through knowledge, persuasion, decision, implementation, and confirmation as they discover, evaluate, adopt, adapt, or reject procedures packaged as SKILL.md files. A demonstration challenges agents with different tools, permissions, and operating constraints to debug and repair broken builds. Agents use basic development tools to clone repositories, run make, and execute tests. Neo4j provides the shared layer for publishing and discovering skills, exchanging recommendations, and recording decisions and outcomes. Some skills improve results; others introduce errors or prove unsuitable for particular environments. The graph serves as both a resource agents consult and an observable history of how practices spread. We distinguish exposure (media) from adoption and adoption from demonstrated (works for neighbor) improvement. Attendees will leave with a practical graph model and example queries for tracing skill discovery, influence, adaptation, and performance; and an approach to testing whether collective learning produces better fixes or simply spreads bad advice faster.