Shrink the surface: governed self-service for analyst-owned dbt pipelines
Liftoff made self-service dbt finally stick for analysts by shrinking what they're responsible for, not by teaching them more of the platform.
Overview
Proposes dbt Mesh to fix failed self-service programs: isolating analysts in a separate AI-native project while the platform controls deployment, config, and CI. Covers using AI agents (Claude skills) to encode dbt/git knowledge and automate scaffolding, tests, PRs, and conventions.
What you'll take away
- 01
Liftoff's fix for self-service wasn't teaching analysts more of the platform, it was narrowing what they're responsible for: a separate dbt-pico project where analysts own only business logic and metric grain, while the platform owns everything else.
- 02
An early version of the AI agent entry point produced too much, 11 files and 620 added lines for one request, so the team consolidated the pattern into a single shared macro, removing 564 lines from the codebase.
- 03
A production dataset that used to take about eight weeks through the engineering queue shipped in under a week once an analyst built it through dbt-pico, and a 30-minute query now returns in under a minute.
Session speaker
Jinze Xin
Liftoff