The practitioners' view.

An Electron shell around the Genie One web app, an Alt+Space launcher the docs never mention, and an updater that could not fetch build 0.2.2 on the day we tested it. What you get on macOS, what it costs in memory, what an admin can and cannot switch off, and who should wait.

The direct engine became the workspace default in July, automatic migration started shipping on 26 August, and the Terraform engine goes away in new CLI releases in September. The nine failure modes worth checking, each traceable to the release that fixed it.

A skill is now a catalog object you grant like a table. What the Beta does, the download-versus-live decision it forces, and why the docs cannot decide what to call it.

Full-text indexes with no embeddings, two SQL functions that do different jobs, and a scaling switch that landed separately. What changed on 1 June 2026 and what you now have to decide.

One account-level screen for data health, AI usage, cost, and tags across your whole Databricks estate: what each page shows, how to turn it on, and what is coming next.

The program guide lists the steps. This is what they were like from the inside, and what mattered most in my case.

Half of it is defined on purpose. The other half assembles itself from dashboards and queries you already have. When the two disagree, a score you cannot audit picks the winner.

Everything the Databricks Certified Associate Developer for Apache Spark exam tests, on one page, in the order the official guide lists the domains.

Databricks gives you a strong engine for enforcing data quality. Writing and maintaining the rules is still manual toil. That gap is where AI earns its place.