They mentioned it in the blog announcement, but totally makes sense

of course part of the motivation was so agents could create graphs

It’s never been easier to by a polyglot, and also paradoxically never as low of a need to be a polyglot

#dataBS #rstats #python

But as data science tooling “grew up,” it borrowed from the business analytics world more and more. ibis, dbplyr, Arrow, ADBC, flight SQL, parquet, Iceberg, Databricks offering more and more sql support, duckdb, and now ggsql represent the triumph of data science tooling.

Data Science is like shadow IT that eventually took over regular IT.

It was long and honestly kinda torturous to have 2-3 versions of the same technology for years, and while data science was growing up kind of worse versions? Eg pandas being slow, or dplyr not supporting in database computation

With the help of the Sandy Hook families, The Onion has reached a long-awaited deal to take over InfoWars.

We've enlisted the help of @timheidecker.bsky.social, who will be InfoWars' Creative Director.

Please stand by for more.

I'm going to be thinking about this for a while -- declarative visualization, in your database engine instead of as an add on. Feels very @evidence.dev but as a SQL extension vs. as markdown?

Is buckets of slop on Facebook. It’s AI overviews. It’s a load of hype that feels like horseshit. Stories of teens committing suicidal at the encouragement of Gen AI. Data centers!

In so many ways, transformative for enterprise and ubiquitous but neutral to bad for most folks in most places today

Listening to @reckless.bsky.social and the folks at the Verge explains a lot.

My own work (output and process) has shifted dramatically in the past year in particular. True for the software engineers I work with, the product folks, the designers, all of it!

But consumers? AI for them ..

It’s called pulling a Bluesky

Friday night: always a good time for some Muppet Treasure Island

I also think the lack of concrete different advice is because we don’t actually know what is optimal for agents. What’s good for humans is better for agents than _nothing_, but I think there’s loads of undiscovered county and I’d bet things look very different in 6 months

GitHub - dagster-io/skills: A collection of AI skills for working with Dagster A collection of AI skills for working with Dagster - dagster-io/skills

Eg, our Dagster skills repo has much more terse instructions on our dbt integration vs. pointing to the human docs, because we found performance worse.

We’re discussing how to handle drift etc

GitHub: github.com/dagster-io/s...

And a blog on lightweight evals: dagster.io/blog/evaluat...

I think the concept of progressive disclosure is the same between agents and humans, but I think the implementations might be different.

We’ve found terse, minimal documentation better for skills (both input and output tokens), which is not true for humans.

Also Crypto is right there!

Pipes are sick, and love open systems that anyone can create and build on

Tough times for Onion headline writers these days when reality is so absurd

Struggling shoe retailer Allbirds makes bizarre pivot from shoes to AI, stock explodes more than 700% Allbirds announced a deal with American Exchange Group to sell its intellectual property and other assets for $39 million in March.

Allbirds: the Microstrategy of AI

www.cnbc.com/2026/04/15/a...

They say never be good at what you don’t want to do more of, not sure how to apply that to reviewing AI code ha

Listening to Wes McKinney on the Test Set, he's got a whole agentic review setup which feels like the eventual future. Humans reviewing AI generated code seems like a mistake of some kind

So cool. Love what you can do with open source and open tools 😎

New post. You built an ML model with {tidymodels} that you want automated in Dagster. Engineering's reply: "Sorry, we need this rewritten in Python to deploy." No more. With the {dagsterpipes} R package, you can run R code in Dagster with full observability. #rstats

joekirincic.com/posts/use-r-...

Everyone social media app needs this

This but also see: people shit talking competitors

The Center Has a Bias Why a measured position on AI tends to lean towards actually trying it.

Beautiful and thought provoking essay by @mitsuhiko.at It’s ostensibly about AI, but it’s a much more general point about useful criticism requires deep engagement, and that most critics don’t want to.

In other words, selection bias rules everything around me.

lucumr.pocoo.org/2026/4/11/th...

Lord grant me the audacity of this cedar playhouse maker who put “add branding plate” as step 54 in a 3 hour setup

I think the job of a software engineer at those AI labs looks different (~no handcrafted code), but the huge returns to great software engineers using AI harnesses to ship makes these talent acquisitions valuable to them

I’ve been absolutely part of the nerds mocking the “it’s too powerful to release” mantra. That feels like the 4 hour erection warning for software. This morning I actually sat down to read details about Mythos. 😳 Um. We’ve got a systemic problem. It’s finding zero days by the dozen. 1/n

This is true if I were a data engineer but since I’m a sales engineer it’s a little harder to convey