Inside dbt Wizard, the AI agent built to show its work
Summary
dbt Wizard, an AI agent for dbt work, validates its own output before any code reaches production. The post features its designer and engineering lead explaining how the agent is built to show its work along the way.
Summary generated by brickster.ai. For the full article, follow the source link above.
More from dbt Labs Blog
How Data Culture measured its way from zero trust to a shippable AI analytics agent
Data Culture built a three-step evaluation framework that took an AI analytics deployment from 0% accuracy to 94.7%. The post walks through how that measurement-first approach established trust and made the agent shippable.
How Verily Health monitors wastewater for 30-plus pathogens with a 5-person team and 1 dbt package
Verily Health's five-person team monitors wastewater for more than 30 pathogens in near real time using a single dbt package. The post shows how a small team can keep broad pathogen surveillance running without a large data engineering footprint.
How Kramp cut analyst onboarding from days to minutes with a self-serve dbt Mesh factory
Kramp's central data team built Data Project Factory, a Terraform module that provisions a GCP project, a dbt project, and three GitHub repos for a new domain from just a team name and owner email, cutting analyst onboarding from days to minutes. Presented at dbt Summit 2026, the setup now supports 25 dbt platform projects and 50 developers across 15 domain teams, with federated governance enforced through CI checks and no inherited superuser access.
How customer pressure pushed dbt past a cautious platform team at Delivery Hero
At Delivery Hero, dbt adoption was driven by customer teams pushing past a cautious platform team, not by a platform mandate. The post traces how that bottom-up demand shaped the rollout and what it means for platform teams weighing how fast to support new tooling.
What Okta learned governing context for AI agents at scale
Okta moved its AI agent context out of the Snowflake UI and into dbt, using a three-layer framework of agent metadata in model YAML, markdown context docs, and semantic views in dbt's semantic layer spec, all shipped through normal pull requests. The context syncs into Agents Schema, so Claude (over MCP) and Snowflake Cortex read the same governed rules, and an analytics engineer fixes a wrong answer by merging one change that every agent inherits.
Beyond medallion architecture: the written rules that keep a growing dbt project consistent
Silja Märdla argues that medallion layers alone don't keep a growing dbt project consistent, and lays out 14 explicit written rules that do. The post shows how codifying those conventions gives teams a shared standard to build against as the project scales.