Summary
Tanmai Gopal, co-founder of PromptQL and creator of the widely-used Hasura GraphQL engine, addresses one of enterprise AI’s most pressing design challenges: how to build a shared “company brain” that intelligently routes knowledge to authorized users without leaking sensitive data across organizational boundaries. The talk draws directly from PromptQL’s deployments at Fortune-ranked banks, Instacart, and AI-native companies.
Gopal opens by showing why naive approaches fail — company brains built in a single burst of enthusiasm inevitably see contribution rates decay over time, because the incentive structure is wrong. His core design principle: each person should own and grow their slice of the company brain organically, rather than having a central team build it wholesale. He illustrates this with a concrete use case — an AI agent automatically drafting an accurate response to a security questionnaire from StitchFix by drawing on knowledge a compliance colleague had previously contributed — and unpacks the access control mechanics that make it both useful and safe.
The talk covers the specific failure modes PromptQL encountered when deploying at banks with strict information barriers, including how scoped access tokens, explicit data governance policies, and agent-initiated access-upgrade requests can satisfy security requirements that would otherwise block deployment entirely. Engineers building enterprise knowledge systems or RAG pipelines for regulated industries will find the lessons from PromptQL’s 15–20 early enterprise partners particularly concrete.
📺 Source: AI Engineer · Published September 03, 2026
🏷️ Format: Deep Dive
