Databricks has moved Genie App Builder into Beta, allowing users to build governed, data-aware apps from natural language prompts inside the Databricks UI. Apps run on a new serverless micro apps runtime with scale-to-zero, are governed via App Spaces and Unity Catalog, and are free during Beta on AWS and Azure.
Databricks quietly moved Genie App Builder into Beta. If you're on Databricks, a workspace admin can enable it under Previews as "Governed agentic app-building". It came out of the Data + AI Summit announcements in June, and it's one of the more interesting agent-for-building-apps releases I've seen from a data platform, so here's a rundown. What it is Genie App Builder is an agent inside the Databricks UI that builds full data apps from natural language. You go to the Build tab, describe the app you want, and it generates a working preview. You refine it with follow-up prompts, then deploy. No code needed, and your data never leaves the platform while you build. The interesting part is that it's data-aware. It knows what's in your Unity Catalog, what your tables and metrics mean, and what permissions apply. So you're not wiring up connections or writing boilerplate. You say "build me a QBR prep app for the sales team with territory filters and an export button" and it builds it against your actual data. Under the hood it generates apps with AppKit, Databricks' TypeScript framework for data and AI apps. How it differs from regular Databricks Apps Regular Databricks Apps: you write the code yourself (Streamlit, Dash, React, whatever), deploy it, and it runs on dedicated compute that's always on and billed while running. Genie App Builder apps: - Built through prompts instead of code - Live inside an App Space, a governance boundary where admins define who can build, what data and resources apps can touch, and how they can be shared - Run on a new Serverless Micro Apps runtime that scales between 0 and 1 instance, and spins down after 30 minutes idle. First request after that has a short cold start - No general internet egress. Outbound is limited to the workspace and Databricks-managed npm/PyPI proxies - In-memory state is lost on scale-down, so anything persistent should go into Lakebase or a Unity Catalog volume During Beta, App Space apps can only be built through Genie App Builder, and Genie App Builder can only build into App Spaces. If you need always-on or horizontal scaling, you still use a classic app. Pricing and availability Free during Beta (App Spaces, the builder and the micro apps runtime). Available in a bunch of AWS and Azure regions. GCP isn't supported yet. You also need partner-powered AI features enabled at account and workspace level. The bigger vision The direction is pretty clear: let non-technical business users build their own governed internal tools, without data teams becoming a bottleneck and without IT losing control. Three pieces fit together: Build fast with Genie App Builder Run cheap with scale-to-zero micro apps, so sporadically used tools (weekly ops briefings, one-off analysis apps) are actually worth deploying Govern centrally with App Spaces and Unity Catalog On top of that, apps can embed Genie Spaces and agents, so many of them are conversational out of the box. And Databricks Marketplace is becoming a distribution channel for apps, so partners can build once and customers install with a click into their own workspace. The examples they showed at Summit: a sales QBR assistant, an ops command center with live KPIs, an exec briefing tool that runs saved questions daily and sends a summary, and a curated reference hub for sales. Curious if anyone here has tried it yet. How good is the generated code once you go past the simple stuff, and how does it compare to the usual vibe coding tools when it comes to working with real enterprise data?
Google is rolling out Sheets canvas, a Gemini-powered feature that turns spreadsheet data into interactive mini-apps via natural language prompts, available globally in English to Google AI Pro and Ultra subscribers.
Discusses the evolution from text-to-SQL to autonomous data agents, comparing custom-built agents using LangGraph with managed platforms like Snowflake Cortex Analyst, Databricks Genie, and PowerBI Copilot.
Databricks introduces Genie Ontology, a self-improving context layer on Unity Catalog that builds a living knowledge graph of business definitions, using OntoRank to resolve conflicts and reduce text-to-SQL hallucinations.
Databricks introduces GPT-5.5 for enterprise agent workflows, achieving state-of-the-art on the OfficeQA Pro benchmark with a 46% error reduction over GPT-5.4.