Introducing dbt Assist: a copilot to accelerate dbt development

dbt Assist — now in beta — lets you quickly generate documentation and tests in dbt Cloud with the help of AI.

Analytics engineering, at its heart, is about solving problems at the intersection of people and data. Crucially, it’s about moving up the stack: taking care of problems that used to require repetitive manual work for you so that you can focus on delivering even more value to the business.

At the annual dbt Cloud Launch Showcase last week we rolled out a new set of tools for your toolbox that does exactly this: dbt Assist (now in beta; visit here if you’re interested in joining). dbt Assist is a set of AI-enabled workflows for common tasks in dbt. It helps you accelerate (not automate) them, allowing you to ultimately get even more done in less time.

dbt Assist allows you to scaffold critical, but time consuming parts of the dbt workflow using AI:

• Writing documentation
• Generating data tests
• Creating Semantic Models and Metrics (coming soon!)

Lets see it in action

Writing documentation

The point of data is for people (and perhaps increasingly LLMs) to be able to use it — and to do that they need to understand it. By documenting your dbt project, you’re setting it up to be useful to your collaborators, be they human or machine. dbt Assist will generate a first pass attempt at documenting your models, which you can then augment with your own knowledge about your data.

Generating data tests

dbt data tests help you build confidence in your data. dbt Assist automatically generates a baseline set of tests for your data, making it easier than ever to ensure test coverage across your entire DAG.

Creating semantic models and metrics (coming soon)

The dbt Semantic Layer allows you to standardize your business definitions so that you can query your most important data across any system and get consistent answers. Investing in the Semantic Layer allows you to develop conversational analytics built on top of it, which we see as a critical interface for LLM systems.

dbt Assist will soon allow you to get up and running with the Semantic Layer by helping build out a first pass of the semantic models and metrics that power the Semantic Layer. Expect to see support for this in the near future!

It’s important to note that as with virtually all AI products, there is an important caveat you should keep in mind: outputs should be verified and validated before you add them to your projects. This is a tool meant to speed up the scaffolding of your dbt projects, and we’ve found internally that it does do that, but it is not a replacement for knowing your data or performing code reviews.

dbt 🤝 AI

This is our first AI-enabled release in dbt Cloud. There are a few guiding principles that we’ve followed in the process of building dbt Assist. We aim to deliver AI-enabled development experiences that are:

• Tailored to the analytics engineering workflow. The first iteration of dbt Assist is a “Swiss Army Knife” style approach — a number of specialized, vetted tools that accelerate specific parts of the dbt development workflow.
• Built on dbt best practices. We’ve baked knowledge of dbt best practices into dbt Assist. That means the code you’re generating is based off of our most up-to-date learnings on how to get the most out of your dbt projects.
• Context-aware. When using dbt Assist, we dynamically include the relevant metadata about your project in the generation so that the system has access to all of the intelligence that you have put into your projects. The usefulness of LLMs is bounded by the context they have available to them, and dbt Cloud is the place to go to get context about your dbt projects.

Going forward, you can anticipate our AI offerings in dbt Cloud will continue to grow. Expect new and more advanced tooling: what you see in dbt Assist today is is the beginning of our efforts here. For a longer-term view of how AI is going to unlock value for data practitioners, read Tristan’s recent thoughts on how dbt Assist + other workflows will shape analytics work moving forward.

dbt is about solving hard problems once, and then being able to move up the stack.

In the olden days, testing and documentation were manual processes managed on an ad-hoc basis.

Just like dbt has helped minimize manual work in other areas, dbt Assist lets you quickly create a first pass of tests and documentation, so that you can focus on your most important work, and spend less time writing YAML.

This is the first step in our vision toward an AI-assisted development experience in dbt Cloud. We’re looking forward to seeing what it helps you accomplish.