Opinion insight on the new announcements on the Databricks Data + AI Summit 2026, written as we attend the event in San Francisco, with an honest and open view as experienced and reflected by our element61 consultants present.
Learnings during the Databricks Data + AI Summit in San Francisco
"A lot is being announced at the Databricks Data + AI Summit 2026, and like everyone else, we are still getting to know these new platforms, products, and capabilities in practice. Below a honest first-impression perspective on what I’m genuinely into, and one I’m not convinced about."
✅ 1. Databricks has a sharp vision on apps
Databricks is making app development on top of enterprise data feel radically easier, while still keeping governance at the center.
First, they gave you an easy way to build apps backed by a real database: Lakebase, their serverless Postgres offering. Now they’ve shipped Genie App Builder, basically their own Lovable / v0.dev, except it intimately understands the platform it runs on and integrates with all your data and ontology layer.
What I love is that they’re actively defusing the biggest footgun of letting business users code:
- With Unity Catalog, platform admins can centrally pre-configure row- and object-level security. The app connects to your data on behalf of the user, so people can’t leak data by mistake. Another reason to really step up your Unity Catalog game.
- Auth is built in. The folks who never think about SSO get it for free, by default.
Honestly, from my own experience, this is massive. The platform makes it easy to protect users from themselves.
Pair that with Serverless Micro Apps that scale to zero, spin up in a couple of seconds, and shut off when idle, and “app sprawl” no longer means going broke.
I like that Databricks has a bold, opinionated vision on the future of software and apps, and that it’s actually backed by their platform.
🤖 2. Sandboxed compute in Agent Bricks
I’m an agent nerd, so this one got me.
Governed compute. Python interpreters. A place to run bash and Python scripts. All inside your own cloud environment.
These past months, I landed on the view that the last real problem for running autonomous agents everywhere, and not just in your terminal or chat UI 😬, is giving them secure serverless compute.
Options like Daytona existed, but for a lot of enterprises, moving data and code off-platform was a non-starter.
Now that Databricks has built this in, I think it makes them a credible end-to-end agent provider.
🤔 One thing I’m not sold on: Omnigent
As a heavy opencode user, I already have a single harness with access to all models, so I don’t really feel the need to orchestrate multiple harnesses.
The supervisor pattern and routing are already doable from the CLI, and there’s a desktop and web app too.
I do like the Unity AI Gateway integration. And to be fair, it governs and routes external, non-Databricks models too, not just models hosted on Databricks.
But here’s my take on cost: the surefire way to cut enterprise AI spend isn’t paying per token by consuming models through the platform.
For power users, it’s enterprise subscriptions straight from the providers. For example, a €200/month plan at OpenAI or Anthropic can deliver thousands of euros of equivalent API usage.
Final thought
All in all, very exciting stuff from Databricks.
```