Data and AI Summit 2025 Roundup

Databricks announced some big news during their keynotes, including a $100 million investment in training and a new Free Edition of Databricks, serverless GPU options, easily deployed AI agents with Agent Bricks, and general availability of Databricks Apps for streamlined app deployment in the Lakehouse.

Let’s go over these announcements in more detail and some implications for our financial services folks.

Databricks Free Tier, Free Training, and University Partnerships

Databricks is launching a new type of Databricks workspace to provide you with all of Databricks' features. The Free Edition of Databricks extends the features available in the Community Edition to give access to AI and ML features. For those of you still using the Community Edition, now is a good time to switch; otherwise, you have until the end of the year before the Community Edition accounts are disabled.

Why is Databricks launching this expanded Free Edition? They are quoting a shortage in skilled labor during the wave of changes that most companies expect from Generative AI. On top of the Free edition, they are also investing in partnerships with universities with their University Alliance Program and general free access to the Databricks Academy (you’ll need to register for the Free Databricks account and then complete your profile when you log in for the first time to the Databricks Academy.

We’re going to test out the new features announced during DAIS 2025 with the new Databricks Free tier to see how far we can take it!

Agent Bricks: Auto-Optimized Agents Using Your Data

Agent Bricks promises to make delivering AI agents, aligned with your data, quickly–”...Days, not Months”. We decided to put this new feature to the test and share what we discovered.

The first thing we noticed is that Agent Bricks is not available in the new Free Edition, so we moved into our (Premium tier)sandbox environment to test Agent Bricks.

The guided experience with Agent Bricks was quick. We got started with some structured data that was available and built out an information extractor within a few minutes–checkmark for being fast. Next, we tested it with a document, Boost GenAi ROI with AI Agents. This was a PDF, which was not currently supported as a source document, but as soon as I tried Databricks triggered a separate workflow to ingest the PDF and generate a text document from it and we were on our way. This workflow created a new volume for parsed files, parsed_<your source volume>. Once tha parsing was done, we moved to adding the new volume as the source and ran our Information Extraction Agent.

This was a simple process that allowed us to create an AI Agent in a few minutes, with a lot of guardrails. This also offered additional tools to test the AI Agent through a SQL warehouse to validate the outputs for a given input as well as creating an AI Agent MLFlow serving endpoint. We’re going to be cost conscious here, so we deleted it. As expected, all of the resources created for the Agent were deleted, except of course the parsed files.

Note: We did want to test out the limits of ML in the free tier, so we ran through these docs for Authoring AI Agents in code. A few notes here, you’ll be able to run a subset of the models available in non-free tiers and LLM as judge was not enabled for this tier. Either way, we were able to get through quite a bit of the agent creation process in the Free Tier through code.

Databricks Apps are GA

Databricks Apps are now GA! On top of being GA, they have also expanded their allowed frameworks to JavaScript frameworks for more robust user experiences. This was allowed in the Databricks Free Tier, limited to 1 App. They offer several starter templates, including the one we used, Gradio, which sets up your application in your users’ folder by default. The apps come baked in with SSO, powered by Unity Catalog for managing users access.

Was it easy to set them up? Yes. That’s it, we clicked the + New button in the Databricks UI, selected app, used the Gradio template, created the App, then deployed it. Again, within minutes.

Why would you want this? To rapidly create tools for your teams, or customers, that are governed, integrated with SSO, and don’t require a heavy lift from operations team to build and maintain.

Side note, Apps are now also sharable through Delta Sharing and the Databricks Marketplace. Databricks ahs come a long way with the Marketplace and they’ve expanded since the days we first implemented Marketplace Listings with our customers over a year ago. It’s exciting to see how this offering continues to grow.

The next updates (Lakebase and Serverless GPU changes) were not currently available in the Databricks Free tier, so you’ll need to have a Premium or Enterprise tier to test these out.

Lakebase: Powered by Neon

You may have heard of the recent Neon acquisition by Databricks. During DAIS 2025, Databricks Co-Founder Reynold Xin brought up that Databricks was a key investor in Neon and worked closely with them well before the acquisition to ensure they had close integration with Databricks. It’s no surprise Lakebase features went from beta to public preview soon after the acquisition. You can now go into your Databricks workspace and spin up a Postgres Database in the Compute > Databases page (assuming your Databricks admin has enabled this feature).

Outside of just having a serverless postgres database you can integrate with your AI Agents and Apps, you’re also able to sync your Unity Catalog tables to a Postgres Database, read only, to meet the needs of your application. This process was quick, first we started a Database, then selected a Unity Catalog table to sync to that database, and after a minute we had our data accessible in Postgres, along with the expected Data Cataloguing and other UC Governance features.

The big selling point of the Neon technology that powers Lakebase is their serverless capabilities, which means you’ll only pay for the storage used and the compute required to handle the incoming requests. Once you no longer need the database, it will automatically scale to 0, eliminating additional compute costs.

Serverless GPU

This is a small update, but an impactful one. You can now select GPU as a compute option for your serverless workloads. For a notebook, you can select the notebook environment option, change the accelerator and you’ll be ready to test out GPU workloads from your notebook.

This will help you quickly prototype your AI/ML workloads within a serverless powered notebook for faster iterations.

Between free hands-on training, next-generation tools for AI agents, applications running on the Lakehouse, and operational databases, DAIS 2025 underscored Databricks’ push to make enterprise-grade data intelligence and AI accessible for everyone. For financial services teams balancing compliance, real-time risk models, and talent shortages, these investments from Databricks lower the barrier to experimentation, making it easier than ever to deliver faster, safer, and more cost-effective deployments.

Bye for now!

Stay tuned for our others posts digging deeper into the new GenAI Features, Clean Rooms, and our personal favorite sessions from DAIS 2025.