01 · Automation Anywhere · 2023–present
From Risk to Reason
Every enterprise wanted generative AI inside their automations. Almost none were allowed to have it. This is how we changed the answer from "too risky" to "approved", and what happened after.
Role: Lead Product Designer. Strategy, end-to-end design, cross-functional leadership
The problem
By 2023 the models were capable enough. The trust wasn't. Legal, security, and risk teams kept blocking AI pilots for the same reason: an automation that acts wrongly acts wrongly at machine speed, thousands of times, before anyone notices. "Add AI" was easy. "Add AI we can answer for" was the actual brief.
The decision
The easy version of this product was a prompt box bolted onto the automation editor. We made the opposite bet: govern first, then generate. Make the safety controls the product itself, not the paperwork around it.
That decision became a three-role contract. Admins decide which models are allowed and watch every skill's behavior in production. Pro developers package prompts into AI skills: versioned, tested, reusable, with quality ratings attached. Citizen developers drag those skills into automations without ever touching a prompt. Nobody gets raw model access. Everybody gets AI.
How I got there
Four directions at once: interviews with internal experts who had watched AI pilots die in review, teardowns of UiPath and Power Automate, cross-functional workshops on what the platform could actually enforce, and a long read of the developer forum to hear the fear in users' own words.
The design
The skill builder is a test bench, not a text box. You write the prompt next to real input data, run it, and rate what comes back for accuracy, tone, and toxicity before the skill can ship. Monitoring isn't a separate audit tool; every skill in production reports usage and quality back to the people who approved it.
What happened
AI Studio became the company's biggest growth engine: 65% of new enterprise sales, 1.6M+ automations powered, and teams reaching first value 3× faster. The approve, create, consume pattern held up well enough that it now underpins the company's agent platform.
Along the way I mentored four designers on the patterns this work produced, and the mentoring outgrew the design team: 350+ colleagues across the company have now been through my sessions on using AI to make sharper decisions. The short version hasn't changed: show the model's work, make review cheap, and never let "it's AI" excuse an unexplained outcome.
