Brickster Newsletter
|
|
01
Inside Databricks Marketplace. We sorted the store’s apps and notebooks into categories, with two charts. What sells is not what is stocked.
|
|
02
What happened this week. Genie grew up, the safety rails got the attention, and the plumbing quietly improved.
|
|
03
From Brickster. Why a catalog full of migration tools and a week about guardrails are the same story.
|
|
|
01
|
🛒 Inside Databricks Marketplace
|
|
Databricks Marketplace · The catalog, sorted
What sells is not what is stocked
We browsed Databricks Marketplace, the store built into the workspace, and sorted its catalog. Supply and demand point in different directions.
|
|
First, the place itself. Databricks Marketplace is the store built into the workspace. It sits in the left menu, next to your data. Providers publish data sets, notebooks, and apps there. You pick something, and it lands in your own workspace. There is no checkout. Free listings arrive on a click, and paid ones start a conversation with the vendor.
We browsed it inside a real workspace and sorted the whole catalog. The first chart below shows the store at a glance. It is a data store, overwhelmingly. The software shelves (notebooks and apps) are a thin annex at the side. Worth knowing before anyone goes in expecting an app store.
The notebook shelf is the vendors talking. They publish worked examples where buyers are stuck and budgets are moving. The biggest cluster is help escaping older systems. Healthcare runs deepest. Data quality and advertising follow. Read the shelf as a map of where the industry actually is.
The apps shelf splits into rough thirds, shown in the second chart. One third runs the platform itself: costs, catalogs, quality checks, descriptions. One third is AI and agent apps, mostly published by Databricks. The last third is identity tooling for the advertising industry.
Then there is what people actually reach for, and it is neither. Near the top of the popularity ranking sit a bridge to Excel and two synthetic data listings. Synthetic data means realistic stand-in records, with nobody’s real details in them. People want their data in the spreadsheet they already live in. And they want safe stand-in data to build and test with, because governance locks away the real thing. That is honest product feedback, found in a store.
|
|
|
What Databricks Marketplace actually carries
| Data tables |
|
| Files |
|
| Notebooks |
|
| AI models |
|
| Apps |
|
| Partner Connect |
|
| SaaS listings |
|
Listings per shelf. Our count from a live workspace, August 18, 2026. Orange: the shelves this story is about.
|
|
|
The catalog in four findings
|
|
What you get
A notebook here is a settled argument
A listing is somebody’s solved problem, written down. You adapt it. Nobody hands you a running system. The value is skipping a month of design debate. The cost is the time your team spends adapting it, so budget that time or do not start.
|
|
|
What vendors publish
Escape routes and deep verticals
The biggest notebook cluster helps buyers leave SAS, Teradata, and Informatica behind. If you are mid migration, you have plenty of company. Healthcare is the deepest subject by far. Vendors put effort where budgets are moving, which makes the shelf a fair map of the industry’s real agenda.
|
|
|
What people reach for
Spreadsheets and safe stand-in data
An Excel bridge outranks most of the vertical accelerators. Synthetic data, realistic stand-in records with nobody’s real details in them, appears twice near the top. Both answer daily frustrations. People want data in the tool they already use. And they need safe records to build with, because governance locks away the real thing. If that surprises your platform team, it is worth a conversation.
|
|
|
Before you click
Three checks, one owner
Check when a listing was last updated. The ranking ignores age, and we found popular listings untouched for months. Check the Access field: some land instantly, others need a request and a wait. For apps, name an owner before the install, so nothing keeps running after the pilot ends.
|
|
|
What the thirty-one apps actually do
| AI and agent apps |
|
| Identity and ad tech |
|
| Running the platform |
|
| Data sharing links |
|
| Web data |
|
Sorted by hand from every app listing on the shelf, August 18, 2026.
|
|
|
What to take from the store
|
|
1. Go to Marketplace for parts, patterns and platform tooling, not business applications.
|
|
2. Read the shelves as market research. Vendors publish where the money is moving.
|
|
3. If your people want Excel and stand-in data, ask what the platform is missing.
|
|
4. Check the update date and the Access field before planning an afternoon around a listing.
|
|
|
|
|
|
02
|
📰 What happened this week
|
|
|
Here is what actually happened in the Databricks world over the past week, in plain language. Four themes, pulled from everything our pipeline read.
|
|
AI assistants
Genie stopped being a demo and started being something teams run
Genie lets people ask questions of company data in ordinary English. This week was about making it work in production rather than in a showcase. Databricks published guidance on pointing Genie at both tables and documents without losing track of who is allowed to see what. Scottish Water described making its capital investment data conversational. And a community post shared a way to cut Genie off when it reaches a spending limit, a control Databricks’ own budgets feature stops short of.
|
|
|
Safety rails
The rails around AI got more attention this week than the AI itself
Several of the week’s items were about containment rather than capability. Databricks demonstrated its Unity AI Gateway stripping personal data out of a request before any model sees it. It also wrote up Omnigent contextual policies, aimed at a subtle problem: an agent quietly combining several harmless permissions into a leak. Databricks says its Smart Routing sends easier work to cheaper models and cuts cost per task by more than 30 percent.
|
|
|
Tooling
Quiet upgrades that mostly buy your engineers their afternoons back
The Databricks command line tool now matches the Python setup on your own machine, which removes a familiar class of works-on-my-laptop arguments. Asset bundles, the way teams package and ship Databricks projects, now deploy without dragging Terraform state along. That becomes the default on August 26, so it is the one date here worth a calendar entry. The Java and Go kits also picked up newer identity handling.
|
|
|
Data plumbing
Your tables can finally hold images and documents like everything else
A new FILE column type lets you keep images, audio, and documents inside a table rather than parked beside it. That matters because unstructured content has been the awkward exception in most governance setups, governed loosely or not at all. Delta Lake shipped a maintenance release. And declarative pipelines’ bitemporal change capture, the unglamorous art of handling corrections that arrive after the fact, resurfaced in this week’s reading; the feature itself dates to May.
|
|
|
|
|
This issue’s lead came out of an afternoon inside Databricks Marketplace, and it rhymes with the week. The shelves are full of migration tools and platform housekeeping, while the most reached-for listings are a spreadsheet bridge and safe stand-in data. The week’s loudest theme was not a new AI capability either. It was the rails built around one. Both point the same way. Maturity looks boring from the outside. It looks like tooling, testing, and somebody owning the result. That is the job we set ourselves here. We read the whole week so you can skip most of it.
|
|
🛒 Ten minutes on the Marketplace shelves will tell you more about where the industry is heading than most analyst reports. Want a head start on what fits your setup?
The Brickster Assistant searches our full archive and answers with citations.
|
|
|
Enjoying the Brickster Newsletter?
Forward this to a colleague who works with Databricks.
Share this issue ↗
|
|
