Lots of prompt tuning and with big packages / SQL the model does not want to return the full statements, but it looks that it can work very well.

We are safe for AI, LinkedIn still recommends a service desk position for me 🤣

And I still remember that on the Dutch Power BI User Days the crowd did not really respond to this announcement 🤷‍♂️. And I was almost standing and yelling YESSSSSS. So no clue why there was such a big disconnect 🤷‍♂️.

VERSION HISTORY IN POWER BI ONLINE IS AVAILABLE!!!!

I tried to convert a SSIS package to Databrick via o3 model. I can tell you. holy moly macaroni!. It works! It simply works. Needs some prompt tuning, but the basis works. Migration project, here we come!.

Framing is everything:. > We need to hack directly in production to get this garbage working. Versus:. > When needed we have post deployment steps in our release pipeline. Sounds waaaaaaay more professional 🤣.

"Minimal viable architecture". Sometimes I wonder. what are we doing? 🤷‍♂️.

Jay!. But I hope it is better performing than Snowflake, roughly 15% faster with more than 1 million rows, listed in the same accouncement.

But as soon as I manually cluster the table & re-enable the auto-clustering kicks in. At least that is what I saw with a different table. So lets try that again. And a very small table then? It was auto-clustered. What.

Auto-clustering on Databrick is weird. Table with 5.2 billion rows, 20 columns, 100 GB, 2100 files. So not the classic 1 TB border for partitioning. But with clustering there are no lower limits, supposedly. Auto-clustering does not kick in.

Still won 2 years ago a quiz from a sponsor, 10 out of 10 questions correct, no-one else was able to top it that day. And I keep coming up with opportunities to remind our Power BI people that I won 🤣.

Tickets for the Dutch Power BI user days in March 2026 just went for sale and I ordered one for the general Saturday! Jay!. For the precon I really need to have the agenda and see if there is an interesting topic as I am not a Power BI developer, merely an architect. .

So we have a view that is absolute top 5%, but the underlying table is according to Unity Catalog not popular. If I would look at enabling for example auto clustering candidates, would I look at not popular tables first. ? Nope.

Delta parquet

Tomorrow presenting what our data engineering team's vision is on our future within the company. I used @AdvAnalyticsUK vid about "Data Engineering is Dead" as a primer 🤣.

Does someone know where this image is coming from? From a methodology or Databricks own?.

Just realized that I have not seen examples where Lakebase on Databricks is used as a metadata database for ETL frameworks?.

Big acquisition. Or maybe even a merger as if I read the documents correctly we basically doubled our turnover 😱.

Ok ignore that. Something is off 😱

Whoah. Some big news just announced internally. Can't say more about it yet.