Honestly, I should have paid attention to this sooner.
You know that feeling when you build a dashboard, you're proud of it, and then five minutes into the meeting someone says "can you just add one more thing?" And then another thing. And another. Until that clean, simple dashboard you made turns into a monster with 47 charts that confuses everyone, including you.
That's been the reality of BI tools for years. You try to answer one question and end up babysitting a spreadsheet empire forever.
Databricks has been quietly fixing this. And what AI/BI Dashboards looks like today is completely different from even a year ago. Here's what actually changed.
It's already included. No extra cost.
This was the first thing that got my attention. AI/BI is built right into Databricks SQL — no separate license, no new vendor, no extra bill at the end of the month.
That matters more than it sounds. When your BI tool is separate from your data platform, you end up with this annoying gap between where data lives and where dashboards live. You're constantly syncing, extracting, and wondering which number is the "right" number. When it's all in one place, that problem just goes away. Your dashboards pull directly from your actual data, always fresh, always governed.
You can literally just ask your dashboard a question now
This is the feature called Genie, and every time I show it to someone for the first time they say "wait, it does WHAT?"
So here's the old way: a stakeholder wants to know something, they ping the data team, the data team builds a chart, repeat forever. The new way: the stakeholder just types their question in plain English, and Genie figures out the SQL, runs it against your real data, and gives back an actual answer.
Not a guess. Not a made-up number. A real, governed result from your actual tables.
The recent update made it a lot smarter too. It used to feel a little clunky — like it was trying to do two things at once and doing both of them halfway. Now it thinks and writes the query together in one go. The answers come back faster and make more sense.
Your dashboard can now take a guess at what happens next
AI Forecasting is one of those features where you think "this sounds complicated" and then you use it and it takes about three seconds.
Find a line chart that has dates on the bottom axis. Click a button that says "+ Forecast." Done. Databricks draws a projected trend line extending past your current data, based on the patterns it sees — things like seasonality and general direction.
Will it replace a proper forecasting model that a data scientist spent weeks building? No. But will it let a business user see a reasonable "here's where this is probably heading" without waiting two weeks for a ticket to be picked up? Yes, absolutely.
Your dashboards can actually look like your company's dashboards now
Okay, I know color themes sounds like a small cosmetic thing. But if you've ever published a dashboard and had someone immediately ask "can you make it match our branding?" — you know it's not small at all.
Now you can set your color palette and fonts once at the workspace level, and every dashboard your team builds automatically uses them. One setting, done forever. This was apparently the single most requested feature from Databricks customers. Make of that what you will.
You can finally track who broke the dashboard
Dashboards now work with Git folders, exactly like notebooks do. Version history, pull requests, code review, the whole thing.
If someone changes something on a production dashboard and the Monday morning report looks wrong, you can now actually trace what happened and when. If you want to test changes before they go live, you can do that in a proper staging setup. It sounds like an engineering luxury but honestly it's just basic hygiene, and it's good that it's finally here.
Your customers can see your dashboards without needing a Databricks account
This one opens up use cases that weren't really possible before.
If you wanted external people — customers, partners, vendors — to see your data, you either gave them a login (messy, expensive) or sent them a PDF (useless). Now you can embed live dashboards directly into your own applications. They interact with real, up-to-date data. They never touch Databricks. You stay in control through the same governance layer your internal teams use.
Think about customer portals where clients can see their own data. Or partner dashboards. Or vendor reporting. All of it now possible without a separate BI tool.
The little things that quietly make your day better
A filter bar at the top of every dashboard shows you exactly what's currently filtered. Sounds obvious. Was missing. Now it's there.
Ctrl+Z works when you're editing dashboards. I know. I KNOW. It should have always worked. But now it does.
Sankey charts are available natively now — really useful for showing flows, user journeys, where things drop off.
You can send scheduled dashboard snapshots straight to a Microsoft Teams channel — image, PDF, and a link, delivered automatically.
Wide tables let you freeze columns on the left so the important stuff doesn't scroll away.
Where this is all going
Databricks is using the phrase "agentic analytics" to describe where they're headed. The idea is that instead of AI just answering questions, it actually helps you build things. You describe what you want, and it drafts the dashboard. You stay in your dashboard and define a new metric without switching to a different tool. You use APIs to bring conversational analytics into your own apps.
The goal they keep coming back to is: a business analyst with no SQL knowledge should be able to build something real and production-ready without asking the data team for help.
My honest read on all of this
I've been around enough BI tools to be skeptical. Most of them promise to make things easier and end up adding new kinds of complexity instead.
What's different here is that the stuff they've shipped actually solves the annoying parts of the job. Genie handles the endless one-more-question requests. Forecasting answers "what happens next" without a data science project. Git integration answers "who changed this." Embedding answers "our clients need to see this too."
Is it perfect? No. There are still things I'd want to see improved. But when I look at the pace of what's shipped in the past year, it's hard not to be genuinely optimistic about where this is heading.
If you haven't opened AI/BI Dashboards recently, it's worth a fresh look. It's not the same tool it was.
What's your experience been with it? Drop a comment below — especially if you've run into something that surprised you, good or bad.
