Speed Is the Advantage: Real-Time Data and AI Agents Across Global Manufacturer

Every CIO right now has a board asking for AI agents and an engineering team quietly explaining that the data isn't ready.

Vishal Mehra runs IT for a company with fabs on three continents, where data arriving late doesn't produce a bad dashboard — it stops a production line. So before GlobalFoundries put AI and data initiatives into production, he started by subtracting: retiring the legacy change-data-capture stack, including a standalone database that existed only to shuttle data between systems. What replaced it streams equipment, process, and scientific data across the fabs, and now runs a growing portfolio of agents across IT, procurement, and other business functions — with identity, permissions, and audit trails handled once, in Redpanda's Agentic Data Plane, instead of per project.

He's also on his second company running Redpanda, five years in. Alex Gallego is going to ask him plainly why — and what he'd tell Redpanda to fix.

This is a deep dive, not a testimonial. Vishal will name the systems he turned off, the metric his team watches now that they didn't a year ago, where his own estimates were wrong, and the first test he'd run if you're sitting on a streaming estate you're unhappy with.

You'll leave knowing:

• Which systems came out first, and how he made the internal case for retiring them
• What he forced to standardize across regionally-run fabs, and what he let each one keep
• Why real-time is a governance requirement rather than a performance upgrade — and what breaks when an agent acts on data that's minutes old
• What he refused to let agents touch, and why the back office proves the pattern before anything near operations