01Major Unity Catalog update. Three GAs and a Public Preview last week retire the open-vs-governed trade-off
02Taming the Databricks UI. The Consumer Access entitlement (Public Preview) that hides the cockpit from business users
03This Week, In Brief. Agents under UC, the Connect wave, packaged vertical AI
04From Brickster.ai: A public Databricks roadmap, plus three more launches
01
🎯 Inside the GA wave
Unity Catalog just turned the corner Databricks has been promising for two years
Three GAs and one major Public Preview in three days. Read together, they retire the open-vs-governed trade-off.
In three days last week, Databricks shipped three Unity Catalog GAs and one major Public Preview. You no longer have to pick between "open formats" and "rich governance."
What shipped, in order
May 12 · Catalog Commits GA. Open commit standard for safe concurrent writes across engines.
May 12 · Native Lakehouse Sync (Public Preview). Lakebase Postgres data replicates into UC managed tables automatically, with full SCD Type 2 history. Per Databricks' own announcement, this is a Public Preview, not GA.
May 13 · ABAC, column masking, governed tags, data classification GA. The full attribute-based access stack, enforced automatically.
May 14 · Open APIs GA. Apache Spark, Flink, and DuckDB can now read and write to UC managed Delta tables without a Databricks runtime in the middle.
What changed: UC is now open enough that external engines work, AND governed enough that ABAC plus classification plus tag policies enforce automatically. That combination didn't exist before.
What it means
For new architectures anchored on Databricks, the open-vs-governed trade-off is gone.
Snowflake's UC counter (Iceberg plus Polaris) and AWS's (Lake Formation) used to credibly say "we're more open." That gets harder this week.
Catching up means matching from the other direction. That's a bigger lift than catching up on one half.
🔄 The trade-off, before & after
Before: pick one
"Open or governed"
Either:
Open formats, build governance yourself
UC, give up engine optionality
ABAC + classification, roll your own
Sync to Postgres via pipelines you maintain
After: both, default
"Open and governed"
Shipped last week:
Spark / Flink / DuckDB write to UC tables (GA)
ABAC, column masking, classification (GA)
Catalog Commits (GA)
Lakebase → UC sync (Public Preview)
02
📚 Taming the Databricks UI
How to hide the Databricks cockpit from business users
Consumer Access can now be set as the workspace default (Public Preview). Here's why it changes onboarding for non-builders.
If you've ever put a business exec inside a Databricks workspace and watched them try to figure out why a cluster is spinning, you've felt the problem. The platform is built for builders. Notebooks, compute, jobs, ML models, pipelines, every surface is one tab away. For someone who just wants to open a dashboard, it's the cockpit of a 747.
Databricks' answer is persona-based entitlements. The core "Consumer Access" entitlement landed back in 2025. The quality-of-life follow-up is the ability to set Consumer Access as the workspace default for new users. That's currently in Public Preview (docs last updated April 2026, so it's been quietly available for a few weeks). The boring-correct setup finally has a path of least resistance.
Why it matters: access governance isn't only a security conversation anymore. It's also a UX one. And UX is what decides whether your CFO actually opens the dashboard, or pings the data team to forward the screenshot.
🎭 The three entitlements
Manage them under Settings → Identity & Access → Groups → <Select Group> → Entitlements.
🟢 Consumer Access
For business users
Read-only access to:
AI/BI Dashboards
Genie Spaces
Databricks Apps
Plus: Power BI / Tableau can still connect to SQL warehouses for them.
🔵 SQL Access
For data analysts
Adds on top of Consumer:
SQL warehouses
Query history
SQL dashboard authoring
🟣 Workspace Access
For engineers
The full builder UI:
Notebooks and jobs
ML models
Data pipelines
⚠️ Two things to watch
Consumer Access has to be the only entitlement. Per the Databricks docs, users get the Consumer experience only if Consumer Access is their sole entitlement. Add Workspace Access on top and they see the full builder UI instead.
The default users group is a trap. Historically every workspace's default users group automatically granted Workspace and SQL access on its own. So you'd add a CFO with Consumer Access, forget to pull them out of users, and they'd still see the full builder UI. Admins had to strip those defaults by hand.
Default-access-to-Consumer (Public Preview): workspaces can now set Consumer Access as the default for new users directly. The pure least-privilege setup is now also the no-extra-steps setup.
It's a one-checkbox change with outsized impact. Business users stop seeing clusters spin up. Security posture tightens. Onboarding gets shorter. Same checkbox. If you run a Databricks workspace and haven't toggled this on (the new default-access setting is Public Preview, available now), it's the cheapest UX win on the table this quarter.
Three other threads ran through the brickster.ai archive this week, across releases, news, videos, and community Q&A. Full breakdown and item-level reading over at brickster.ai/digest.
AI / Governance
Agents, now under Unity Catalog
Databricks is pushing agent governance into the same plane as data governance. MCP servers can now be registered and governed inside UC. Genie's evolving "skills" interface lets teams scope custom agents to specific data domains. The pattern: every agent gets a UC identity and a policy boundary, by default.
HubSpot's Lakeflow Connect integration hit GA this week (Jira's been in Beta since January). The Google Sheets connector itself went GA back in April, but a follow-up landed this week: run history of scheduled refreshes, viewable from inside the sheet. And Native Lakehouse Sync entered Public Preview, replicating Lakebase Postgres into UC managed tables with full SCD Type 2 history. The pattern: every popular SaaS surface gets a managed UC ingestion path instead of being a Fivetran problem.
Three packaged AI applications shipped, all built on Databricks Custom Agents and Mosaic AI. PipelineIQ for forward-looking sales forecasting. Predictive Quality for manufacturing defect detection. Genie for Merchandise Intelligence for retail markdown timing. The platform-vs-apps line is moving.
53 features status-tracked from public sources, with the source post linked from every claim.
The most-asked question about any Databricks feature has always been: is it GA yet? Databricks doesn't publish a public roadmap. Their release-notes surface is scattered across blog, docs, and keynote videos. So we built one. brickster.ai/roadmap consolidates every feature we track into one status-stamped view.
53 features bootstrapped. 40 GA, 9 announced, 4 in preview. Browse it as a tree (Unity Catalog has 21 sub-features hanging under it), filter by status or category, click any node for the source post and the status history (announced → preview → GA) with date stamps on every transition.
Honest about its limits: every entry traces to a public Databricks post. We don't see the private AE-shared roadmap. If a feature exists but hasn't been blogged, it won't be here. Free, no signup, click any feature to verify the source in one tap.
Every public input we read, grouped by stream. News feeds, GitHub release repos, YouTube channels, the Data + AI Summit archive, community forums. Each with a live last-seen timestamp. Forty inputs, no private feeds, no scraped paywalls.
A live bubble cloud of every topic, sized by mentions across the archive (releases, news, videos, projects, community Q&A). Greener bubbles are trending up vs the previous period; redder ones are cooling off. Toggle the window (today, this week, this month, all time) and click any bubble for the source breakdown. One image, one read of where the ecosystem's attention sits this week.
brickster.ai/mcp · the entire Databricks archive, inside your AI client
The MCP server. Drop one JSON snippet into Claude Desktop's config, restart, done. Your AI client now has typed tools over the entire brickster.ai archive: semantic search, recent releases, recent news, recent videos, curated reading list. Same setup works in Cursor, Codex, Continue, any MCP-aware client.
🤖 Stay ahead of the Databricks curve
Ask the Brickster Assistant about anything in the archive. Cited answers, no hallucinations.