Whatever you're building in Metabase, there's an AI path that can help you get there faster: Metabot, MCP, the CLI, or agents. This session is the decision map — for each thing you're actually trying to do, which one fits the job and why.

Agents are great until one rewrites something it shouldn't. In this workshop you'll set up Workspaces as a guardrail, giving an agent room to do real work while keeping it contained, reversible, and reviewable. Bring a laptop — you'll configure the containment patterns yourself, and see the failure modes that make them necessary.

Bring a laptop and your own messy data, or use a dataset we'll provide. You'll go from raw sources in many shapes and places to modeled, trustworthy data and a working semantic layer — live, in the room, with AI handling the tedious parts. We'll show you where it saves hours and where you still want a human in the loop.

As more of your stack gets automated and more decisions get made by models you didn't write, the things open source gives you — transparency, control, no lock-in, your data staying yours — stop being nice-to-haves. The argument for why being open matters more in the age of AI, not less.

Everyone's experimenting with AI on their data. What's less clear is what's holding up in production. We'll open with concrete examples of where Metabot, MCP, and agents are delivering value today and where they're still rough, then open the floor so you can compare notes with people working on the same problems.

You don't need to be a frontend engineer to build a custom visualization anymore — you need a clear idea and an AI pair. We'll go from "I wish Metabase had a chart that…" to a working custom viz, live, including where the model gets you there fast and where you'll still want to write it yourself.

Bring a laptop and build a working data app, live: a custom layout with your own querying interface and visualizations, all running on verifiably correct data from Metabase's semantic layer. You'll leave with something running and a feel for where the numbers hold up when someone acts on them.

Bring a laptop and an app. With an AI pair, you'll embed Metabase live — working through guest, modular, and SDK embedding as you build, and solving the problems that come up along the way. Leave with Metabase running inside your own app and a sense of which embedding approach fits which job.

Row- and column-level access and sandboxing were already subtle. Now an agent can query on someone's behalf, and "who can see this?" needs a new answer. We'll walk through the permissions model and its sharp edges so you can defend your setup when the requester is code.

What running Metabase looks like at scale: complex permissions and security, embedded and multi-tenant analytics, org-wide governance, and hard-won patterns like filtering questions before they hit a shared dashboard.

Where slowness actually comes from, what caching can and can't fix, how to use model persistence and transforms to precompute the expensive work, and how to read a slow query and find the real bottleneck.

A room for people running Metabase on their own infrastructure, whether that's the open source version or a paid plan. We'll kick off with the operational realities — upgrade strategy across a dozen releases a year, scaling, the gotchas — then open the floor so you can compare notes with people running the same setup.

Not a lecture. We'll open with onboarding and adoption patterns that have worked across technical and non-technical teams, including now that there's an AI assistant in the mix, then turn it to the room: what's worked for you, what hasn't, and how you get people to self-serve.

"Can't AI just do this now?" is the question every data team is getting. How to quantify your impact and usage, tie analytics to decisions people actually made, and defend the investment in terms that survive a budget meeting.

Every product makes choices that puzzle people. This is the philosophy behind Metabase's: why some things are deliberately simple, why we said no to features you asked for, and the design principles underneath.