Descriptions:
In this Y Combinator interview, GigaML founder Varun describes turning down a $550,000 offer from a leading quant firm to build AI agents for customer support — a decision his peers thought was irrational at the time. GigaML now powers support infrastructure for DoorDash, one of the top three US crypto exchanges, and multiple Fortune 500 companies, with the system pushing deflection rates from the traditional 10–15% toward a 90–95% target using fully conversational AI.
The conversation covers GigaML’s technical approach in surprisingly concrete terms: the central lever is iteratively improving markdown policy files to move business KPIs like resolution rate and CSAT. Varun describes this as a generalizable pattern applicable beyond customer support — to compliance, ITSM, and internal support — and credits it as the insight behind their rapid enterprise expansion. Landing DoorDash as a customer when GigaML was just eight people required a three-month pilot with zero downtime plus a warm YC introduction to DoorDash leadership.
The interview also includes candid career advice for college students and early engineers, a frank account of GigaML’s YC interview (where the original edtech idea was rejected on the spot), and Varun’s view on why the current LLM moment is analogous to earlier platform shifts. For anyone tracking how AI agent startups are winning enterprise contracts and what the actual product architecture looks like under the hood, this conversation offers more operational detail than most founder profiles.
📺 Source: Y Combinator · Published May 29, 2026
🏷️ Format: Interview
