dbt Summit 2026: the keynotes and product sessions

There’s something we keep hearing from data teams. A chief data officer at a health insurance administrator framed it well recently. He has two data engineers he wants to fast-track onto AI work. He has executive support. He has budget. And he still can’t start.

“Everybody's making noise around AI. Can you do something around AI? And I'm thinking, ‘But your data is still not at the level where you can put an AI agent on top of that.’”

He already knows what most organizations are about to find out the hard way. It’s a gap this year’s dbt Summit content happens to address.

dbt Summit lands September 15-18 at The Cosmopolitan in Las Vegas.

You have done this before

Ten years ago, dbt changed what it meant to be a data professional. The analytics engineer emerged, with production-grade pipelines at speed and scale, governed data, and real influence over decisions across the business.

It’s easy to forget how that felt at the time. The shift was disorienting. Plenty of people weren’t sure where they’d land. It turned out to be one of the best things to happen to careers in this field.

We’re at that moment again. AI is rewriting how data gets used, and it’s raising the stakes on every data decision made in the next 18 months. The consumers of data are shifting from analysts in a BI tool to agents working continuously, at machine speed, without a human checking every answer.

The teams who thrive will be the ones who level up, which is why you need to be at dbt Summit this year.

Two keynotes, four pillars

The dbt Summit keynote is built on four pillars, and they line up closely with what we hear directly from data teams and what they are up against.

Level up the engine with faster parsing, better scalability, and real cost control, wherever you run dbt.

Level up for AI and what your agents consume. Structured, governed context is the missing piece between your warehouse and an agent whose answers you’d stake a decision on.

Level up with AI and what you consume with dbt Wizard and the agentic workflows around it, so you build and ship faster while keeping control of quality and spend.

Level up the stack with an open, flexible infrastructure. Your stack stays yours.

Thursday brings the Community Keynote with Grace Goheen and Jeremy Cohen. This is a celebration of the dbt community, how dbt Core and the engine unite under one framework, and how the work continues in the open.

Go deeper in the product sessions

The keynotes are the headline. The product breakouts and roundtables are where you find out how it all works, straight from the product managers who built it.

Optimizing your runs for lower compute, fresher data, and faster iteration with dbt State. Most dbt projects rebuild the same models every run, whether anything changed or not. dbt State checks warehouse metadata and model SQL, works out whether the result would actually change, and then builds, skips, clones, or auto-defers to production. Average compute savings run 30%. Reuben McCreanor walks through freshness SLAs, auto-cloning from prod, and the orchestration logic you get to retire, whether you run dbt Core, the dbt platform, or something in between.

dbt Wizard: your AI teammate for data development. dbt Wizard is the coding agent built for data, not software. Ani Venkateshwaran and Brandon Thomson walk through its three modes—develop, analyze, and discover—the agent skills framework that makes it extensible, and a live demo of the fully autonomous analytics engineering workflows dbt’s own team has built on top of it.

Rebuilding dbt Core in the open: faster runtime, adapters, and docs v2. Years of dbt Core v1.x left teams with a Python runtime that slowed on big projects, fragmented adapters, and a docs experience that couldn’t keep up. Hope Watson unpacks the rebuild: a Rust-based Apache 2 engine, a cleaner adapter model on ADBC and the Arrow ecosystem, and parquet-backed artifacts. You’ll see side-by-side parse speed demos and get a real migration path from v1.x through the v1.12 parser on-ramp.

The semantic layer is dead. Long live the semantic layer! Semantic layers used to be a solved problem you built years ago for dashboards. Then agents showed up. Zach Mandell makes the case that your semantic layer is now the most load-bearing piece of your AI strategy, shows where MetricFlow and the dbt Semantic Layer are heading, and covers how the dbt MCP server turns governed metrics into context your agents can use. This is the session that answers the hallucination problem directly.

Product roundtable: open data infrastructure in practice. Jack Lowery and Anna Lee host a working conversation on unlocking dbt projects with Apache Iceberg and mesh patterns. Roundtables are small and discussion-first, so come with the problem you’re actually stuck on.

Apache Ossie: Realizing Semantic Layer Portability. Define a metric in one tool, redefine it in the next, and your dashboards, notebooks, and agents all end up disagreeing on what “active customer” means. Apache Ossie, the vendor-neutral, open-source semantic interchange spec co-led by dbt Labs, Snowflake, Salesforce, BlackRock, and RelationalAI, changes that contract. You’ll see how MetricFlow, now open-sourced under Apache 2.0, compiles governed metrics into a portable format so BI, conversational analytics, and AI agents finally agree on the same business logic.

The next wave of data infrastructure is in the Lake. The modern data stack got you here, but it won’t get you to where agents need to operate. Russell Christopher and Casey Karst lay out Open Data Infrastructure: land data in open formats like Iceberg in storage you control, then pick the right compute for any job and swap tools without rebuilding pipelines. You’ll leave with a framework for building open, portable, AI-ready infrastructure with Fivetran + dbt.

We said I do. Now, we’re joined at the DAG. Where did this data come from, what happened to it, and who’s using it? Roxi Pourzand shows how Fivetran and dbt unify ingestion, transformation, and consumption into a single lineage graph—one that falls out of the systems doing the work instead of being stitched together after the fact—unlocking real-cost visibility tied to usage, PII that stays governed downstream, and agents that reason with the full picture.

With great context comes great autonomy: leveling up your agent context. An agent querying your warehouse shouldn’t be guessing what “active” means or which revenue table to trust. Ben Moser and Kevin Kim demo how dbt structured context and the Fivetran context layer turn raw metadata, governed semantic models, and real usage into the agent schema that powers accurate results—and how to let agents reason beyond fixed definitions when the question calls for it.

More sessions are landing between now and September. Browse the full agenda to build the rest of your week.

Come find out what you get to build next

The organizations that win the AI era won’t be the ones with the best models. They’ll be the ones whose data foundation was ready when it mattered. That foundation already has a name, and you built it.

dbt Summit 2026
September 15-18, 2026
The Cosmopolitan, Las Vegas

Your next level starts here. Register for dbt Summit.