How UiPath's CTO is building the unsexy infrastructure that makes AI actually work

During his keynote at UiPath's Fusion conference, Chief Technology Officer (CTO) Raghu Malpani doesn't showcase the latest breakthrough model or promise autonomous everything.

Instead, the demonstrations focus on testing frameworks, evaluation scores, and governance layers for a healthcare discharge process. The emphasis is on simulations, execution traces, and Personally Indentifiable Information (PII) detection – unglamorous infrastructure for making AI agents work reliably in production.

When I sit down with Malpani afterward, I ask why this focus on what most vendors treat as afterthoughts. His answer is blunt:

AI agents are easy to demo. Making them production-reliable is hard. That's why so many AI projects fail to deliver.

That assessment frames everything UiPath is building. Across the keynote and our interview, the pattern becomes clear – an organization positioning itself around the premise that enterprise AI success depends less on model superiority and more on solving problems like orchestration, governance ...