Customer support · Production

Vercel's internal data agent d0 democratizes analytics access across the company

The problem

Running AI agents in production requires expertise in security, observability, reliability, and cost optimization that is rare even as building becomes easier, creating a growing shadow IT problem. Vercel's own data access was previously limited to professional analysts, leaving engineers, marketers, and executives unable to self-serve answers from the data warehouse.

Workflow diagram · grounded in source
1
User asks question in Slack
trigger
“A user asks a question in Slack: "What was our Enterprise ARR last quarter?" d0 receives the message, determines the right level of data access based on the permissions of the user, and starts the agent workflow.”
2
Permission-based data access routing
routing
“determines the right level of data access based on the permissions of the user”
3
Semantic layer exploration
ai_action
“The agent explores a semantic layer: The semantic layer is a file system of 5 layers of YAML-based configs that describe our data warehouse, our metrics, our products, and our operations.”
4
AI SDK model calls
ai_action
“AI SDK handles the model calls: Streaming responses, tool use, and structured outputs all work out of the box.”
5
Durable step orchestration
integration
“Agent steps are orchestrated durably: If a step fails (Snowflake timeout, model hiccup), Vercel Workflows handles retries and state recovery automatically.”
6
Isolated sandbox execution
ai_action
“Automated actions are executed in isolation: File exploration, SQL generation, and query execution all happen in a secure Vercel Sandbox.”
7
AI Gateway model routing
routing
“Multiple models are used to balance cost and accuracy: AI Gateway routes simple requests to fast models and complex analysis to Claude Opus, all in one code base.”
8
Answer delivered in Slack
output
“The answer arrives in Slack: formatted results, often with a chart or Google Sheet link, are delivered back to the Slack using the AI SDK Chatbot primitive.”
Reported outcome

One person built d0 in a few weeks using 20% of their time, where the same work would have normally taken months of engineering effort.
d0 now gives engineers, marketers, and executives self-serve natural-language access to the data warehouse. Vercel's customer support agent handles 87% of initial questions autonomously, and a lead qualification agent lets one SDR do the work of 10.

Reported metrics
Customer support initial questions handled autonomously87%
D0 agent build effortone person built d0 in a few weeks using 20% of their time
Estimated traditional engineering effort for comparable buildmonths of engineering effort
Reported stack
SandboxesFluid computeAI GatewayVercel WorkflowsAI SDKSlackSnowflakeClaude OpusPython
Source
https://vercel.com/blog/anyone-can-build-agents-but-it-takes-a-platform-to-run-them
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

One person built d0 in a few weeks using 20% of their time, where the same work would have normally taken months of engineering effort.

What tools did this team use?

Sandboxes, Fluid compute, AI Gateway, Vercel Workflows, AI SDK, Slack, Snowflake, Claude Opus, Python.

What results were reported?

Customer support initial questions handled autonomously: 87%; D0 agent build effort: one person built d0 in a few weeks using 20% of their time; Estimated traditional engineering effort for comparable build: months of engineering effort (source-reported, not independently verified).

How is this customer support AI workflow structured?

User asks question in Slack → Permission-based data access routing → Semantic layer exploration → AI SDK model calls → Durable step orchestration → Isolated sandbox execution → AI Gateway model routing → Answer delivered in Slack.

Related customer support cases