dbt Copilot's 2024 debut became dbt Wizard by 2026

The setup

dbt Labs unveiled dbt Copilot on 8 October 2024, according to the company's own announcement at its Coalesce conference. The post describes an AI engine embedded within dbt Cloud, launched in beta, aimed at documenting SQL models, drafting semantic-layer definitions, and writing baseline data tests, work the post frames as necessary but tedious, done manually up to that point. Retrieving the vendor's current product page on 16 September 2026 at the same web address now returns a different name, dbt Wizard, installed as a CLI, a desktop app in private beta, or a hosted workspace.

What the documents show

The 2024 announcement states dbt Copilot could automatically generate documentation, semantic models and data tests, plus a chat feature for querying defined metrics, all described as beta capabilities dbt Labs built, not independently verified output. The 2026 page describes broader tasks under the Wizard name: refactoring a model and updating every referencing file, tracing a broken pipeline to a root cause, and migrating models with a reviewable diff before anything saves. Both pages share the same premise, that grounding an assistant in a project's own lineage, tests and compiled state produces more reliable suggestions than a generic assistant without that context.

The friction

Both documents state, in close to identical terms two years apart, that a person stays in the approval loop. The 2024 post says dbt Copilot keeps you in control, with a human always in the loop. The 2026 page says every change is auditable and nothing ships without review. The current page also describes a self-validates step, the tool running its own checks before presenting a change, a vendor claim about reducing what a reviewer must catch, not that review is no longer needed. Customer quotations citing incident-reduction figures are self-reported by named customers in dbt Labs' own marketing material, not an independent benchmark, and are not restated here as fact.

What changed in the work

What the two documents together support is continuity of one commitment, that nothing merges without a person approving it, across a rename and a substantial feature expansion, from a documentation-and-testing helper to a multi-interface agent that also refactors and migrates code. That the product needed a new name and three new interfaces within two years is itself evidence the original beta scope did not stay fixed; a team that adopted dbt Copilot for its original three tasks should not assume today's dbt Wizard is the same tool with a new coat of paint.

• Which of the documented tasks, refactor, build, investigate or migrate, is this workspace actually using, versus what it was set up for originally?
• Does the self-validates step run against this team's own tests, or only against dbt Wizard's general checks?
• Are the customer-reported incident-reduction figures on the current page from a setup comparable to this one?

Two announcements, two years and one rename apart, agreeing on exactly one guardrail is a reasonable signal that the guardrail is the part dbt Labs considers non-negotiable.