How Vercel removed 80% of their agent’s tools with d0
Lessons learned from operationalizing a Slack-based data analyst agent
Discover how Vercel’s engineering team improved the accuracy of their internal data agent, d0, by doing the unthinkable: removing almost all of its tools.
Join Abhi Sivasailam, VP of Data at Vercel, as he breaks down the architecture of d0. You will see how we moved away from brittle, individual function calls for every metric and instead empowered the Vercel AI SDK to query our Snowflake data warehouse directly using a read-only schema.
You'll learn:
Why giving LLMs too many tools leads to confusion and how to fix it.
How to safely expose your data warehouse to an LLM.
How we stopped writing tools for every new KPI and created a self-serving agent that adapts automatically.
See how shifting from "tool-heavy" to "context-heavy" design can transform your internal AI agents into critical operational assets.
Speakers
Abhi Sivasailam
VP, Data
