Scaling AI Fluency

In the previous unit, you explored how Zapier launched its AI journey before consensus existed — and how bundling the CEO mandate with immediate enablers turned a polarizing moment into a productive one. But a launch moment, no matter how well-designed, only gets you started. The harder question is how you sustain momentum across hundreds of people over months and years without building a parallel universe of AI training that competes with the work everyone's already doing. That's exactly the challenge Brandon Sammut tackled next — and the principle he landed on was deceptively simple: stop creating new programs, and start embedding AI into the rituals your organization already runs.

As Brandon described, Zapier's guiding principle for scaling AI fluency was to "integrate AI learning and experimentation into as many of Zapier's existing practices as possible." The reasoning was practical: "there's already a lot going on in the business" and the goal was to "minimize the overhead or complexity" of building AI muscle. Hack weeks became AI hack weeks. Onboarding became an AI fluency onramp. The work itself became the curriculum.

One of the highest-value, lowest-effort moves was creating a dedicated Q&A channel — Slack or Teams — where "anyone can ask any question about anything related to AI" with "no question too big or small." Critically, Zapier didn't rely on goodwill. They assigned AI power users — early adopters who'd been experimenting since GPT-3.5 — and told them that roughly "10% of your bandwidth" each week was now formally allocated to giving "quick helpful answers" in that channel. Brandon noted this was meaningful for the power users too: their leadership was "calling me into helping my peers move through the FUD, scale their AI fluency." The formula was simple — a single place for any question, plus dedicated helpers — and as Brandon put it, "it can't get any easier operationally than that."

Building on those embedded rituals, Zapier eventually realized they needed an explicit standard for what AI fluency looked like — particularly as new people joined. Brandon described creating and then open-sourcing an AI fluency framework with three levels: capable, adoptive, and transformative. At the capable level, candidates have "dabbled with AI" — even in their personal lives — and demonstrate genuine curiosity. That's the minimum hiring bar for most roles. At the adoptive level, people are "starting to produce measurable impact" through AI in their daily work — the threshold for more AI-intensive product roles. The transformative level is intentionally stretchy: these are rare individuals "helping their organizations rethink the work itself" — not just improving existing processes but redesigning "what's for humans, what's for AI, role cards, team design."

This framework didn't just inform hiring — it reshaped onboarding. Zapier redesigned their program as a 90-day arc with "a lot of AI fluency skill-building within it." And the results of this sustained, embedded approach showed clearly: the most recent engagement survey measured AI usage in core work at 97% across the company. Perhaps most telling, the first two departments to hit 100% were "the finance team and the people team — not the engineering team, not the product team."