AI Adoption & Skills Framework
The company had decided AI was worth investing in. The harder question was how to get real uptake without it going sideways — because AI amplifies whatever you point it at.
The challenge
Left unmanaged, people build their own prompts, reach for whatever tool is nearest, produce wildly varying output, and duplicate each other's work without knowing it.
The approach
Rather than write a policy, I built distribution. I mapped every persona in the business — executive, sales, delivery, project management — and then mapped what each actually does: the tools they work in, the technologies they deliver on, the work that fills their week. That mapping came from evidence rather than assumption, drawing on the internal code repository to see which languages and projects people genuinely work on, and project management data to cover the work that doesn't produce code.
Against each persona I inventoried what capability already existed: native M365 integrations, Microsoft's Power Platform and Fabric plugins, skills available in the Anthropic marketplace. Where nothing existed, I built it.
Architecture
Everything is held in a GitHub repository as the single source of truth, versioned and reviewed before distribution. That repository generates a managed manifest mapping personas to skill sets, which Claude Code picks up and deploys into each person's environment.
The part worth pointing to isn't the distribution mechanism, though — it's what the internal skills actually encode. A number of them aren't filling a marketplace gap at all; they're the practice's own working principles made operational. How we approach design, how we make decisions on evidence rather than assumption, how we test before we ship — those are skills now, not just things people are told. And underneath all of it sits a helper file layer: anything learned on a project, technical or contextual, gets captured and routed to by the skills that need it.
Outcome
Nobody has to work out which capability to reach for. Someone working in delivery on Power Platform is already enrolled in what that role needs, along with the global standards. New starters inherit the right baseline on day one rather than discovering it over months. And the whole thing compounds — it gets better with every project, rather than staying fixed at whatever it was the day it launched.