Shorts: The AI Maturity Arc
A brief framework to measure adoption
I worry with taking advantage of AI there’s a “moment in time” to almost everything we learn. Right now I’m looking at AI adoption through my current lens of a large skill/agent library. I’m in a different spot from someone still figuring out their agent instruction file. I see it like this:
No library: the primary agent (Claude, Gemini, etc) improvises everything with behavioral and task context both living in the prompt. This is fragile and inconsistent.
Emerging agent library: subagents exist but skills are thin. Context is buried inside agent instruction files, invisible by default. Iteration is blind.
Mature library: behavioral context lives in skills, agents, and a well-curated primary instruction file. The primary agent passes only task context. Invocation prompts get shorter as the library gets better.
This isn't a strict progression. If you know the shape of the library you’re building toward you can skip ahead (hint: start working on skills). At stage three the primary agent gets out of its own way. Then subagents start performing much closer to how you actually expect.