By Michael Domanic, Section Head of AI

A little over a year ago, while I was still the Head of AI at UserTesting, I watched something happen to AI adoption in the company that interested me.

OpenAI launched the 4o image generator, and within days our AI Slack channels were flooded with Studio Ghibli versions of people's dogs, their kids, their weekend trips, and our CEO (sorry Eric!). It was fun to see, but it felt like a meme moment that would burn itself out in two weeks.

But that’s not actually what happened.

Within two weeks of the launch, we saw roughly a 15-20% spike in sustained active users across our ChatGPT deployment. And it was a spike that never saw a correction. When I dug into the numbers, the picture got more interesting.

A significant portion of that spike came from people who had been using ChatGPT once a month at most, or had never touched it at all. They came in for the Ghibli pictures. But then they started using chat. They started using the custom GPTs that had been built and shared across the workforce. Play time converted them into real, regular users of AI.

There was a second pattern in the data. The people who were already active users and also played with the image generator started expanding into features they'd never touched: Deep Research, reasoning models, custom GPTs. Their mental model of what the tool could do had shifted, and they explored accordingly.

For the dormant users, play got rid of a major barrier: the energy it takes to start using AI. The image generator required zero prior knowledge. They didn't need to understand how to prompt, or what a model was, or why any of this mattered. They just wanted a Ghibli version of their dog. Once someone crossed that threshold, the tool stopped feeling foreign.

For the active users, play broke an existing frame. They'd been thinking of ChatGPT as a text tool. The image generator proved otherwise, and a broken frame made them more receptive to re-examining their other assumptions.

Brice Challamel, formerly the Head of AI at Moderna and now an exec at OpenAI, talked about this dynamic at Section's AI:ROI Conference last year. His argument was that play is how every species with a capacity for evolution learns - kittens, puppies, baby humans. Give people a safe space to experiment and get weird with a new technology, and they will likely surprise you. I saw the data version of that principle at UserTesting, and it permanently changed how I think about driving adoption.

Given this insight, I started putting more focus on public spotlights for people doing interesting experiments, even the ones that seemed frivolous, and encouraged others to try the same. Someone seeing a colleague's Ghibli dog and laughing is a more effective onboarding moment than any training deck. Social contagion is a real adoption mechanism, and the good news - it’s free.

I'm running a similar playbook now at Section. I'm encouraging people across the company to experiment with building things in Claude Code. That includes people who aren't in technical roles and have never opened a terminal. The goal is not to turn everyone into a developer. The goal is to break the frame, lower the floor, and let people discover what the tool can actually do when they're not being told what to use it for.

The organizations I've seen struggle with adoption have one thing in common: they invested only in the serious use cases. They skipped play entirely and went straight to ROI. In doing so, they missed the fact that play is how you build the population of people capable of delivering the ROI in the first place.

See you next week, 

Michael
Your fellow Head of AI

‍