Descriptions:
Garrett Galow, head of product at WorkOS — the enterprise authentication platform powering Cursor, Anthropic, and OpenAI — presents a conference talk at the AI Engineer summit on how his team built an internal AI analytics tool called Studio. The core problem Studio solves is the slow, manual loop of non-technical employees asking engineers for custom SQL queries to answer one-off business questions.
Studio is built on LangGraph for agent orchestration, using Claude Opus as the underlying LLM, with integrations into Snowflake for data warehousing, Linear for project management, and Notion for documentation. Galow gives a live demo showing Studio autonomously identify which blog content drives the most new customer sign-ups, running multi-step Snowflake queries, self-correcting on errors, and even patching a visual UI bug in its own output. A second demo shows support teams using a self-serve Slack bot to look up individual customer session histories without engineering involvement.
The talk distills three hard-won implementation lessons: sequencing (run pre-flight checks and ask clarifying questions before touching data sources), guidance layers (explicit rules for how the agent should interact with each integration), and schema comprehension (teaching the agent to navigate complex database structures reliably). The result is a tool WorkOS’s go-to-market and support teams use daily to build and share dashboards autonomously — a practical blueprint for any engineering team looking to reduce internal data bottlenecks with AI agents.
📺 Source: AI Engineer · Published June 11, 2026
🏷️ Format: Hands On Build
