Coalesce 2025: Rewriting the future of data, analytics, and AI

Coalesce 2025 is off and running with exciting keynotes, lively hallway conversations, and more than 100 breakout sessions on AI, Fusion, Iceberg, and everything in between. More than 2,000 people packed the general session this morning at Resorts World in Las Vegas, with another 10,000 joining online from around the world, making this the largest Coalesce event to date.

In a jam-packed 90 minute keynote, dbt executives, product leaders, and some of our customers shared the latest developer experience features of the dbt Fusion engine, demonstrated how to reduce your cloud bill with state-aware orchestration, and showed off the dbt platform’s new governed agentic AI experiences.

But today wasn’t just about new platform features. It was about giving practitioners and leaders the tools they need to move faster, cut costs, and trust AI in ways that weren’t possible before. In other words, we are rewriting your expectations for how data work gets done.

The era of open data infrastructure

On Monday, we announced a merger with Fivetran to set the standard for open data infrastructure .

The topic of open data and the merger of two complementary technologies led today’s keynote. While there is still lots to sort out over the coming months, dbt Labs and Fivetran have committed to the following:

• dbt Core and Fusion will both continue to be shipped under their current licenses.
• We will continue to maintain dbt Core indefinitely.
• We will continue to support and foster the dbt Community

Be sure to check out a replay of the opening keynote to hear from both Tristan and George about why we're excited for this next step, together.

Now, on to the product announcements! 👇

Rewrite the developer experience with Fusion

Faster dev cycles, fewer errors.

We’ve rebuilt dbt from the ground up. The new Fusion engine is available in Preview in the dbt platform for eligible projects. Back in August, we launched Fusion into Preview for local development in the CLI and VS Code extension. With today’s release, you can choose to run your projects on Fusion whether you develop locally or in the cloud.

Built in Rust, Fusion parses 30x faster than dbt Core and introduces a compiler and stateful architecture that understands SQL deeply across platforms. That foundation unlocks key features that data developers have been asking for including:

• Intellisense, hover insights, go-to-definitions, and instant refactoring
• Live previews of CTEs without leaving your flow
• dbt compare, a brand-new way to see data diffs directly in VS Code

Fusion is also compatible with dbt Core and the projects you’ve built in the platform. Your existing dbt code will largely just work with it, and we’ve built Autofix tools to help smooth any upgrade issues. dbt Core isn’t going away—it’ll continue to be supported indefinitely—but Fusion is where innovation is headed.

Adoption is strong, with over 1,500 weekly average projects running on Fusion and over 8,000 weekly average users of the dbt VS Code extension. With today’s release, many more teams will get the chance to run their projects on Fusion for the first time.

Click here to take Fusion for a test drive.

Rewrite the rules with state-aware orchestration

Run when it matters. Reuse when it doesn’t.

Also available in Preview for projects running on Fusion in the dbt platform is state-aware orchestration, a new way to avoid unnecessary costs associated with executing data pipelines while still meeting your data business requirements.

Here’s how it works: instead of rebuilding every model in your DAG, Fusion has the stateful intelligence to automatically skip the ones that don’t need to be refreshed—because either no data has changed upstream, or you’ve defined certain rules for how often you want your models refreshed—reducing wasted compute and delivering faster pipelines.

Automatically reuse unchanged models, save 10%

Turning state-aware orchestration on is simple. Once your projects are running on Fusion, just flip the toggle in the dbt platform, and dbt automatically pinpoints which models should be refreshed, based on whether or not new data is produced upstream. Models without any upstream changes are automatically reused.

Results from our beta cohorts showed this simple change resulted in an average of 10% cost reduction in data platform costs, with no required changes to existing projects.

Tuned configurations for more savings

This is just the start. Instead of wrestling with schedules, we have made it easy to fine-tune your configurations: you can declare your freshness requirements, and then dbt will detect exactly which upstream tables have new or changed data and update models accordingly based on the freshness targets you’ve set across your project or on each model.

For example, you can set your maximum staleness window before a model needs to rebuild (say, six hours) with a build_after=6h configuration. Or you can define whether all sources need to be refreshed before updating a downstream model (updates_on=all).

In our beta cohorts, tuned configurations reduced annual cloud costs by at least an additional 15%+.

More efficient testing

Testing also gets smarter with Fusion. New column-aware and aggregated testing features eliminate redundant checks so you’re not validating the same data over and over. Early testing shows these improvements reduce CI and testing costs by an average of 4%.

Run Fusion in production and save 30%

Put together, these optimizations can drive a potential 29%+ reduction in annual data costs:

• 10% from just turning on state-aware orchestration
• 15%+ from tuning the configuration with business SLAs
• 4% from more efficient testing and CI

Intelligence is now the default in dbt. You don’t need to rebuild your project: just run it on Fusion.

Learn more about state-aware orchestration here and see how dbt Labs saved 64% on our own projects!

Rewrite how you work with data with AI

Governed, auditable AI in the tools you already use.

Before we get into how AI works with your data, let’s talk about how you interact with your data, particularly as it relates to development and query. dbt Canvas, generally available (GA) in May 2025, makes it easy for analysts to contribute safely to production pipelines in a visual way. New to dbt Canvas is the ability to upload a CSV file and drag it into your project.

Another popular platform feature, dbt Insights, is now generally available today, giving anyone governed, fast analysis with confidence.

AI-powered analytics is here

AI has moved fast in the past two years. What started with basic SQL generation has become end-to-end project automation, and dbt sits at the center by powering your AI with a structured context layer—models, metrics, tests, and lineage—so automation is trustworthy and results are reliable.

Today at Coalesce, we announced several major updates to the dbt MCP server, the universal bridge between AI tools and dbt structured context layer:

• Local MCP now supports OAuth, so you can securely log in with your dbt credentials.
• The remote MCP server is now generally available and exposes one secure endpoint per environment, making it simple for tools like OpenAI, Anthropic, and Cursor to connect to your team’s dbt projects.
• Fusion MCP tools are now integrated with the dbt MCP server, giving agents richer SQL comprehension and lower-cost execution.

On top of this foundation, we introduced dbt Agents: out-of-the-box, auditable agents that extend governance into everyday AI workflows. The following are coming soon to dbt:

• Developer agent (coming soon) : Explains model logic, predicts downstream impact, validates before merge, and can draft or refactor from prompts. It will run in VS Code or dbt Studio and is powered by dbt context, so every change can be shipped quickly and safely.
• Observability agent (coming soon): Helps you monitor jobs, pinpoint likely root causes, and cut resolution time. Designed to reduce noise and cut investigation time dramatically.
• Discovery agent (beta): Helps you find the right dataset or metric in plain language, along with clear definitions and why it’s trustworthy. Surfaces governed sources and dbt lineage so anyone can explore data with confidence.
• Analyst agent in dbt Insights (beta) answers natural language questions with governance intact. Answers come with definitions, lineage, and tests—what we call “answers with receipts.”
• Sign up for the dbt Agent waitlist today.

Developer agent (coming soon)

Analyst agent (beta)

The impact is already visible. At Norway’s sovereign wealth fund, NBIM, any employee can now use conversational analytics powered by the dbt MCP server and Anthropic’s Claude to ask questions and get verified, accurate results.

With dbt MCP and dbt Agents, AI is no longer a side experiment. It’s how governed data gets built, managed, and consumed at scale.

Finally, we announced that MetricFlow is now open source under Apache 2.0, with co-maintenance from partners like Snowflake and Salesforce, and aligned to the Open Semantic Interchange (OSI) ecosystem efforts . This makes MetricFlow a reliable engine and critical AI infrastructure that ensures metrics like “revenue” or “active users” are consistent across dashboards, notebooks, and AI agents. Open sourcing MetricFlow enables partners and the community to build together on a a common, open engine to accelerate AI adoption and preserve trust in every result.

Rewriting how data work gets done

Day one of Coalesce 2025 was about rewriting your expectations for how data work gets done.

• Rewrite the developer experience with Fusion, now available in Preview on the dbt platform for eligible projects, as well as in Preview for local development.
• Rewrite the rules for your cloud costs with state-aware orchestration, also available in Preview for dbt platform projects running on Fusion.
• Rewrite how you build, manage, and consume data in the age of AI with dbt Insights and the dbt Remote MCP Server, both GA today and with new AI agents in beta and coming soon.

This is more than a roadmap. It’s already here. Thousands of teams are adopting Fusion today, and the results are clear: better performance, lower costs, and AI they can trust.

If you’re ready for a deeper dive into dbt Fusion, check out our upcoming webinar, Speed, simplicity, cost savings: Experience the dbt Fusion engine.