Most of what we see about "using AI at work" is polished. Demos, tips, before-and-afters. What's missing is how you got there: which prompt you tried first, where you had to backtrack, when you chose not to use AI at all.
That's a problem because tools change too fast to train on specific features. Training on "how to use ChatGPT" becomes obsolete in months. Past technology shifts taught us this: technical ability is just the beginning. What matters is changing how we think and work to make the most of new tools.
So at Zapier, we tried something different. We asked people to show us their messy process working with AI, not their polished results.
Table of contents:
AI fluency is about how we think and work
Building AI fluency isn't just about tool training. It's about showing people what skilled AI use looks like in practice: the decisions, the iteration, the judgment calls.
What behaviors show up in the best AI users? A few patterns we watch for:
Starting from the outcome, not the tool. "What am I trying to do?" comes before "Let me use AI."
Actively steering. Iterating across multiple prompts, confidently rejecting bad output, refining until it's right.
Knowing when not to use AI. Recognizing when something is too sensitive or nuanced, when human judgment matters more.
Using AI to stretch thinking, not replace it. Asking it to challenge your ideas or explore unfamiliar angles, not just produce a first draft you clean up.
Owning the final decision. The output is input. You bring the context, empathy, and judgment AI can't have.
What we built: Behind the Build
We launched an internal series called "Behind the Build." Here's how it works.
I asked volunteers to record themselves solving a real problem with AI. I gave them questions to guide them, like: what they were trying to do, why they chose AI, where things got complicated, and where they brought their own judgment.
The recordings were raw. Dead ends, course corrections, moments where they argued with the AI or scrapped output entirely. We edited them down into posts that show what they did, then added short text sections pointing out the learning moments.
One example: our teammate Emily was building a talent sourcing tool. She asked the AI to search for real candidates. It kept saying it was doing exactly that. She clicked the links. The profiles didn't exist. She pushed back multiple times before the AI finally surfaced the actual limitation.
What that reveals: she wasn't accepting claims at face value. She tested the output, noticed the gap between what was promised and what was delivered, and kept pushing until she understood the real constraint. That's skilled use.
Why showing the process works
I've noticed that direct peer learning is so important for learning AI. There's no substitute for seeing someone else click around, like watching a chef prepare a meal (not just the end result). That's where you see the choices, the corrections, the judgment calls.
The messy middle is where the learning happens. When to push. When to pivot. When to stop using AI and bring your own judgment.
What you might try
If you're building AI fluency at your company, here's what's working for us:
Try a version of this:
Ask 3-4 people to record themselves solving a real problem with AI (5-15 minutes of raw footage).
Make it low stakes. Emphasize you're learning from the process, not judging the outcome.
Point them toward a specific task so it's easier to compare approaches.
Edit it down, and share what you learned with the team.
What to look for when you watch:
Do they start from the outcome or jump straight to using AI?
Do they iterate and refine or accept the first output?
Do they bring their own judgment, context, and empathy into the work?
Do they know when to override or stop using AI?
The goal isn't to create polished content. It's to make the invisible visible—so people can learn from how their peers actually work with AI, not just what they built.
Most companies approach AI fluency by teaching tool skills: how to prompt, which buttons to press. That's useful, but what actually transfers across tools is how you think, work, and decide in partnership with AI. Show the messy middle, not just polished outcomes.
Related reading:
