Paving the Road to Production: Governed Enterprise AI Agents — Harish Gaggar, Credit Karma

Register an agent through an internal portal and receive more than a repository: the deployment also needs identity, memory, traces, evaluations and secret handling. Harish Gaggar walks through that paved path for enterprise agents at Intuit Credit Karma. Local development progresses through container builds, review, security scanning and Kubernetes deployment, with the supporting resources provisioned consistently. The architecture places a centralized MCP gateway between agents and tools so authentication and routing are part of every request. Custom MCP servers expose reviewed tools, while OpenTelemetry, Tempo, Prometheus and Grafana make failures, latency and access behavior visible. Gaggar then follows a LangGraph workflow from interpretation and planning through evidence gathering, policy checks and human approval for sensitive actions. Threads, runs and checkpoints preserve state, while separate tracking tables record model use, tool calls, costs and evaluation fields. Demonstrations inspect those records rather than establish a validated quality score for the whole system. A PII filtering layer aims to intercept sensitive inputs before they reach models, memory or traces; detection and redaction remain controls to evaluate, rather than a guarantee that no secret can escape. The final component prunes, reranks, deduplicates and compresses context. Its examples show large token reductions, including about 90% in a demonstrated query, without a paired answer quality study or measured business ROI. The practical point is to build the surrounding framework from the start, giving each agent a consistent route through access control, observability, durable state and accountable action. Speaker info: - https://www.evidyafoundation.com/about-us Timestamps: 00:00 - From experiments to enterprise agents 1:28 - Missing production foundations 2:54 - The agent and supporting frameworks 4:06 - Registration, local development and the paved path 6:27 - Review, containers and Kubernetes deployment 8:14 - Reviewed custom MCP servers 9:47 - MCP gateway authentication and routing 11:48 - Logs, traces and gateway metrics 14:01 - LangGraph policies and human approval 16:59 - Threads, runs and memory checkpoints 19:48 - Agent tracking and evaluation records 23:27 - Filtering sensitive inputs 24:25 - Context optimization and token reduction demos 27:33 - Build the framework around every agent