Technology adoption is a people problem dressed as a tooling problem. I have seen the dressing fall off enough times to stop being fooled by it.
When Generative AI arrived in the enterprise conversation, the emotional range was wide — excitement, skepticism, outright fear. Some engineers worried about replacement. Some managers wanted magic by Friday. Both reactions, in my view, misunderstood the moment.
This was a skill transition, not a headcount verdict.
Training that respects practitioners
We trained more than 200 engineers — not death-by-slide-deck, but hands-on work: prompt patterns that fail in production, RAG designs that respect document structure, agent workflows with observability and rollback baked in.
The goal was practitioner confidence. I wanted an engineer on a Wednesday afternoon to say, “I know how to build this responsibly,” and be right.
The guild model
Hackathons spark. Guilds sustain. We stood up an AI guild so knowledge traveled across distributed teams — office hours, shared libraries, postmortems, honest stories about what broke.
Guilds reduce the lottery of “which team got the one person who read the docs.” They turn individual learning into organizational memory.
From threat to amplified expertise
The cultural shift I cared about most was subtle but measurable: teams stopped asking whether AI would replace them and started asking how it could make their best people faster.
That aligns with everything I believe about expertise. Senior domain knowledge is more valuable in the AI era, not less — if leadership protects it and tools amplify it instead of flattening it into generic answers.
Production pairing over performative demos
Demos are cats on skateboards — fun, forgettable. We paired on production deployments until teams internalized the unglamorous parts: tracing, evaluation, cost controls, access policies, incident paths.
Several hackathon projects graduated because they arrived with operators, not just presenters.
Expert Vision angle
Training emphasized how experts retrieve and synthesize information — not linear reading, but structured foraging. That cognitive lens improved RAG designs and interface choices in ways generic “AI 101” curricula miss.
Open question for leaders
Are you giving your teams a credible learning loop — practice, feedback, reuse — or a mandate and a ChatGPT tab?
The difference shows up in retention, incident rates, and whether AI becomes a capability or an anxiety. I choose capability. The work above is how we pursued it.