From backend PR to dbt model: The merge request with no human in the loop

Every time your engineering team ships a feature, your data team plays catch-up. Schema changes, new tables, renamed columns. Someone has to track them down, write the models, run dbt, open the merge request.

Simba automates that entire loop. Built at Pet Media Group, Simba is an autonomous analytics engineering agent that listens to GitLab webhooks, detects backend schema changes, researches business context, writes production-ready dbt models across every layer, runs the model in its own sandbox environment, and opens a reviewed, tested merge request, with no human in the loop.

This session walks through how we built Simba: the architecture, the skills-based instruction system that lets the data team refine agent behavior without touching code, how we minimize token usage, and the hard lessons from running it in production.