Figma shares its team's AI adoption practices: how to turn colleagues from onlookers into actual users?

Start each interaction with a genuine problem.

Select work that colleagues genuinely need to complete in the near future, such as organizing existing materials, revising page copy, or preparing a business overview. Invite participants to bring real inputs and challenges they have encountered, so the demonstration revolves around this actual work. Participants can see what information they need to provide and easily judge whether the examples suit their own tasks.

Turn demos into examples colleagues can repeat

After exchanges, save the task context, publicly shareable example inputs, instructions used, and parts of the output requiring human modification. This allows those who did not attend the demo to know where to start and how to judge if the results are usable. For business materials that cannot be shared, create a separate example that retains the task structure instead of stuffing sensitive content directly into shared records.

In subsequent exchanges, prioritize seeing what happened after colleagues actually used the tools: where background needs to be supplemented, where things are misunderstood, and where modifications were needed to meet requirements. This accumulates specific work experience. Merely recording participation numbers or demo counts is insufficient to show that colleagues have integrated the tools into their work.

Allow trial failures and clarify continuation conditions

Leave room to exit trials: if the burden of organizing inputs and correcting outputs exceeds current methods, pause the trial for that task, switch to a different problem, or supplement with examples. In discussions, share failed inputs and steps unsuitable for the model to help colleagues judge when to continue using it versus when to handle it manually.

Check results with a fixed input

Have a colleague bring a depersonalized weekly-report example with input, prompt, corrections and final draft. Another colleague reproduces it independently next time, recording missing steps and help needed before updating instructions. Attendance and tool openings measure participation, not completion. No team trial or work-time measurement was conducted here.