AI-powered data development: how agentic SDLC & dbt-MCP transformed our data engineering workflow
Zendesk cut model-promotion errors from 20% to zero by pairing Claude Code with the dbt MCP server.
Overview
At Zendesk, moving a dbt model between data layers used to take hours of manual effort, causing redundant logic and reconciliation issues. Covers how agentic SDLC and dbt MCP streamlined model promotion, plus methodology, tools, and key takeaways.
What you'll take away
- 01
Connecting Claude Code to the dbt MCP server gave the agent real project context, lineage, tests, and metadata, instead of making it guess at a 1,000-plus-model project's structure.
- 02
The results were direct: a 20% human error rate in model promotion dropped to zero, and the combined promotion-and-validation cycle time compressed by more than 90%.
- 03
An internal AI Skills Library with review, governance, and evaluation layers is what let an eight-person team effectively double its capacity without adding headcount.
Session speakers
Suhas Jangoan
ZendeskAyan Putatunda
Zendesk