I am simultaneously worried by the dbt/fivetran merger, and quizzical why we, as paid professionals in profit-seeking entities, should have parasocial relationships with open source python packages.

I am still convinced the only thing European clouds need in order to be competitive is an actual IAM solution with passwordless M2M authentication. Stuff like Azure Managed Identity and IMDS. Please throw in Oauth CLI authentication while you’re at it.

Finished a workshop that seemed to have the mantra "why don’t people realize they can use this awesome technology that they don’t know about to solve this problem they don’t know they have?". I guess there’s a lot of money in selling people both the problem and the solution.

I’m ready to call it. Databricks governance is a fucking dumpster fire.

Databricks copilot is amazingly bad. Run bad command > tells me the syntax is X > Get error > tells me the syntax is Y > Get error > tells me the syntax is Z.

Absolutely bizarre that Databricks on Azure (and GCP) only lets you have one unity catalog per region - basically forcing you to develop in production.

Wrote up how to get GitHub Actions with OIDC to Azure working. Don’t know what that is? It is basically a way for your GitHub actions to log in to Azure without using a password. And Microsoft is making it increasingly easy to set up. radbrt.com/posts/github...

Really want @markzandi.bsky.social and the Inside Economics crew to comment on the idea that was floated to only publish employment numbers quarterly. I get the need to have new data ahead of fed meetings, but other than that a lot feels like impatience in a trench coat.

The difficulty of creating a grouped aggregation with a weighted average in Pandas tells you a lot about the users. With R, this is built-in. I wish the pandas aggregations were pluggable and we would be able to create things like weighted mean, gini etc.

💯don’t demo with fake data to the business. Or, actually, either demo with very correct numbers or so obviously fake numbers that everybody gets it.

So, my iPhone lock screen can’t render the letter "a"?

Read @edzitron.com’s guest post (the unpaywalled part): www.wheresyoured.at/the-remarkab... The article is about why companies buy software they don’t need. I have argued almost the exact opposite - that they build unnecessarily complex software themselves. 🧵 1/

This morning I read blog posts that convinced me that there are only two problems in software: that we buy software, and that we build software. I guess there is only one problem in software: software.

If you like my recent «Data Engineering Toolkit» one-pager, you're going to love this one. It's the Data Engineering Toolkit, presented in a long format. This resource is for experienced data engineers or people who want to dive into the DevOps side of DE.

When did my clipboard start syncing between my Mac and my iPhone? As cool as it is… why would anyone want that?

Sane security defaults needs to be included in the documentation and tutorials too. Databricks' use of user tokens is past it's expiration date. And the alternative is hardly documented: radbrt.com/posts/databr...

Damn, the new baidu models look sweet. Love that they are up on huggingface and has a space to try out.

I decided my new persona will consist of insisting that AI is useless and will also steal people’s jobs.

I miss drunk tweeting so much. Usability is far more important than cool features. Keep it simple. Keep it useable.

Talking to real actual humans that I know in Europe who are canceling their US travel plans because they don’t want to risk detention at the border.

I guess when a website URL has "prod" in the name I should be happy they don’t develop there.

This will be interesting, from the team that has been able to keep abstractions simple.

Several years ago, I asked the Inside Economics podcast w/ @markzandi.bsky.social what BLS statistics they wish existed. I thought they answered an improved version of that question a year ago, but my exact question came up this Friday. The answers were interesting both times. 🧵

Watching the dbt launch showcase and getting strong "we’re the new Informatica now" vibes.

I am starting to suspect SAS is uniquely suited to corrupt the brain of non-progammers. Few languages make it harder to do the right thing. It naturally pulls towards chaos.

Vibe-coding a web app and half of my prompts are because of regressions. On the plus side, I use 3.5 brain cells for this and in a few months my job as a primitive QA engineer will be automated.

Every once in a while I remember a former client that was in discussions with a web analytics company (this was before google analytics). She wanted to mail a brochure to repeat visitors to the site. Our common concept of the internet had solidified a lot over the years.

Wrote about what is and isn't moving in the data space now. Please tell me I'm wrong. radbrt.com/posts/in-the...

211 years today, Norway got its own constitution. May 17 is like The Purge but for patriotism. Take all the everyday patriotic symbols you see over the year in USA, and jam them into a single day. That day is today. I have taken refuge in the mountains.

I just learned that SAS accepts "not=" as inequality operator. Whoever did this chose violence.

TIL: Azure ML can spin up a hosted Posit Workbench, BYOL. All my googling for Enterprise level RStudio SaaS solutions never turned up Azure ML.

Are other people also analyzing means within deciles or is that some weird shit?

I don’t develop in prod, I prod in develop.

@jayatillake.bsky.social do you have any idea what happened with the dataBS crowd? Did everyone just go back to X? Has Trump sucked all the oxygen out of the room, so nobody has brain cycles left? Is it because LLMs? Or is data finally boring again?

"You can outsource the work, not the responsibility" is the main principle to remember with LLMs too.

I once again find myself looking for an RStudio solution that doesn’t require a procurement process. All architectures are procurement driven to some extent. Fabric is procurement driven to the fullest extent possible.

I’m just going to admit it: buying an OLED TV was worth it.

💯. Some years ago it started feeling like half my emotional life was tangled up in a shell script. It was time for a change, in more ways than one.

"I’m not stupid, I’m just unlucky when I think" is my new goto excuse for everything.

Is anyone using RMarkdown in Databricks? Does it work? Can you knit a document without downloading all the stuff locally? #dataBS

Everyone was talking about the spark sql pipe syntax on LinkedIn today, are we fans here on bsky as well or is it just a LinkedIn thing?

Rarely do I see such exquisite nerding over official statistics, especially not from big podcasts like @tracyalloway.bsky.social’s Odd Lots: www.bloomberg.com/news/article...

Interesting piece on COBOL migration (in the context of doge). www.wethebuilders.org/posts/what-i... This one really resonated:

I have, unfortunately, expected Metaplane to be picked up by someone. Now, Datadog has acquired them. There are worse fits. I liked Metaplane for the transparent pricing and simple signup. Hopefully, orgs that use Datadog will have happier data teams now. www.metaplane.dev/blog/metapla...

I hypothesize that every developer in the world would have bought Microsoft a beer for not renaming Azure AD to Entra ID. Microsoft would have been *so* drunk.

It is late (in Europe), and if you are still doomscrolling please stop and watch the capital children’s choir do an a Capella rendition of Crystal Castles’ Untrust Us instead: youtu.be/g7a2u9v0V0c.

I've been exploring how to do Iceberg with the fewest dependencies possible. tl;dr -> @duckdb.org , @amazonwebservices.bsky.social S3 Tables, and a sprinkle of Python is awesome. jakthom.dev/blog/zero-in...

Do you have a favorite tech radar? Do you maintain one yourself? I’m thinking in the lines of the zalando or thoughtworks radar, for data. #dataBS

I have two Bluesky accounts. This one where I have tried to gather all the data people, another one with EU and US political commentary. Ever since the inauguration, I can’t tell the two feeds apart.

The SAS SCL language is the only circle of hell Dante needed. The other 8 are just wasteful spending.