Paving the Road to Production: Governed Enterprise AI Agents — Harish Gaggar, Credit Karma
Enterprise AI agents fail in production due to missing foundation controls: auth gateways, MCP servers, eval systems, and PII protection.
“most of the people are missing the gateway to make sure that agent can, you know, communicate to the gateway to get authenticated first with the right auth providers”
A senior engineer at Intuit Credit Karma outlines the structural reasons AI agents fail in enterprise production, centering on absent foundation controls like authenticated MCP gateways, tracking and evaluation pipelines, memory management, and PII blocking before LLM calls. The talk presents a layered governance framework addressing these gaps, including custom MCP servers, a centralized gateway with auth routing, context optimization for token savings, and agent monitoring. This is a practitioner-level signal that enterprise teams are now systematically codifying what good agent infrastructure looks like beyond the prototype stage.