1,200 analytics requests a day: Clay and Vercel's Bet on AI-Native Analytics
summary
Vercel’s D0 answers over 30,000 questions a month for 800 employees with 45% of queries coming from other apps and agents, built on Eve where agents are treated like a file system; Eve wires tools (Snowflake, ClickHouse, OAuth, HTTP API, Slack, web UI, MCP) into folders and YAML context layers (semantic, metrics, events, terminology, guides). Clay’s Monty, built in under a month by one engineer, saves 40 human hours a day, clones the repo as its KB, and runs real workflows (example: 15 tool calls, six governed queries, one matplotlib chart, 2m12s, zero errors) to catch a softening renewal.
AI-written summary · about
chapters (8)
- 0:00d0: 30,000 questions a month, 45% from agents
- 2:12The minimum lovable data agent
- 5:19Capabilities as folders in Eve
- 7:13D0's five structured context layers
- 8:58Monty: the codebase is the knowledge base
- 13:44A 5 a.m. account health deep dive
- 16:33Why you should build, not buy
- 17:42The next thousand data agents
related talks
19m1,200 analytics requests a day: How we built Vercel's data agent
25mFrom Analytics to Agentic Apps
16mGame Data for All: How AI Is Reinventing Game Analytics
19mAutomating 90% of support tickets: How we built Vercel's support agent
18m3 Oracle AI Use Cases
30mself-serve AI analyst
50mOpening Keynote
16mTransforming Analytics With AI: A Comprehensive Guide to an Advanced Tech Stack