iiinova — where your digital self keeps evolving

Reposted by Jon Harmon (he/him/his)

This week's #TidyTuesday is about hospitals in urban centres, from the EU's JRC. The data has quality problems, so I plotted the counts themselves instead. 95% of the counts are even, which suggests hospitals are counted twice. Code: github.com/gkaramanis/t... #RStats #dataviz

Two charts side by side under the title "Two hospitals, or none*", with a subtitle about data from the EU's Joint Research Centre. On the left, a bar chart of urban centres by their number of hospitals, from 1 to 32. Even numbers have much taller bars than odd ones. Two hospitals is the most common count at 1,982 urban centres, followed by four at 1,080, six at 677 and eight at 379. No odd number has more than 26. There is no bar at 1, marked with an arrow and the text "No urban centre has a single hospital". A second label reads "95% of hospital counts are even". On the right, one horizontal bar per region, with thickness showing how many urban centres the region has. Each bar is split into a blue part for urban centres with a hospital count and a grey part for those without. The blue share is smallest in Central and Southern Asia at 30%, then Sub-Saharan Africa at 47% and Eastern and South-Eastern Asia at 52%. It is largest in Europe at 95%.

071

Jon Harmon (he/him/his) @jonthegeek.com · 02/10/2026

Step 1 toward a happier life: Delete twitter and never think about loading it again.

150

Jon Harmon (he/him/his) @jonthegeek.com · 02/10/2026

I haven't figured out how this fits the metaphor quite yet, but BTW... You can use agentic ai to help prepare you to use agentic ai! But you have to guide it carefully and review, and it will probably often make sense to just do it yourself. Oh, hey, there it is!

000

Jon Harmon (he/him/his) @jonthegeek.com · 02/10/2026

(I don't think I've ever had a talk + over-arching metaphor pop into my head so fully formed 😊)

100

Jon Harmon (he/him/his) @jonthegeek.com · 02/10/2026

If you're adding agentic AI to a project, do some work up front to make sure it's ready for AI. Make sure the existing code has the style and design that you want. And please please please make sure your code is covered by tests (so you and the agent can tell if it's turning into the Red Skull).

231

Jon Harmon (he/him/his) @jonthegeek.com · 02/10/2026

Seeing the difference in how the same agentic AI models in the same IDE perform differently on different projects, it hit me that AI is like the super soldier serum in the MCU: "The serum amplifies everything that is inside, so good becomes great; bad becomes worse."

Stanely Tucci as Dr. Erskine, inventor of the super soldier serum, from Captain America: The First Avenger. He is in a barracks, preparing Steve Rogers mentally for the procedure.

180

Jon Harmon (he/him/his) @jonthegeek.com · 29/09/2026

Definitely not you, don't worry!

010

Jon Harmon (he/him/his) @jonthegeek.com · 28/09/2026

It continues to be disheartening to be the first human to read the code in a PR when I'm supposed to be *reviewing* it. 🙃 That PR has *your* name on it, not Claude's! Have at least a LITTLE pride in your work!

3201

Reposted by Jon Harmon (he/him/his)

Gabby Palomo, PhD @gabspalomo.bsky.social · 28/09/2026

Hope you all enjoy this week's #TidyTuesday dataset. It's a good one! Lots of Qs to explore regarding hospitals and pharmacies proximity to urban centres. Also, now that I think I am getting the hang of it on how to submit a database, I will continue to explore other potential databases. It's fun!

0101

Jon Harmon (he/him/his) @jonthegeek.com · 28/09/2026

I have 5 728-day-old hydroponic tomato plants which I use more for meditation than cultivation. 1 of them (closest in the photo) is barely holding on. I'm trimming the one next to it fairly hard to get the poor little guy some light, but it only has a few small leaves left to try to absorb it 😢

An Aerogarden hydroponic system with 5 tomato plants. The plant closest to the viewer is mostly bare of leaves.

130

Jon Harmon (he/him/his) @jonthegeek.com · 28/09/2026

These could make good #TidyTuesday datasets! dslc-io.github.io/tidytuesdayR...

dslc-io.github.io

Curating a Dataset

010

Jon Harmon (he/him/his) @jonthegeek.com · 28/09/2026

A subset of this could make a good #TidyTuesday dataset! dslc-io.github.io/tidytuesdayR... (or the whole thing, but we'd need it divided up into separate CSVs, each <= 20MB so GitHub doesn't choke)

dslc-io.github.io

Curating a Dataset

110

Jon Harmon (he/him/his) @jonthegeek.com · 28/09/2026

This could make a good #TidyTuesday dataset! dslc-io.github.io/tidytuesdayR...

010

Reposted by Jon Harmon (he/him/his)

Anna Fergusson @annafergusson.bsky.social · 23/09/2026

This is the week in my intro data science course where students learn that runif() is said r-unif not run-if, & they are like "duh" but no one had ever said it out loud before so they didn't know. Def read/watch more about this from @amelia.mn & of course @felienne.mastodon.social.ap.brid.gy, links⬇️

3193

Jon Harmon (he/him/his) @jonthegeek.com · 23/09/2026

Doh! I just noticed that the #TidyTuesday poster actions were disabled! Some people definitely found the datasets, but I'm getting the posts out now...

040

Jon Harmon (he/him/his) @jonthegeek.com · 22/09/2026

This doesn't feel like a thing you do if you expect to have to face voters in free and fair elections: wapo.st/4hmb3QY They've won plenty of elections, so I don't think we can just assume they're out of touch.

wapo.st

Trump officials plan to cut hundreds of thousands from ACA, claiming fraud

Vance targeted ACA enrollment as part of a broader administration crackdown on suspected fraud.

140

Jon Harmon (he/him/his) @jonthegeek.com · 22/09/2026

@masalmon.eu pointed that out, too. I need to move jarl up to the top of my to-learn! I'm sure it'll end up being like pkgdown, which I put off for months before "learning" it well enough in about 5 minutes that I could start using it.

110

Jon Harmon (he/him/his) @jonthegeek.com · 22/09/2026

BTW, coverage standards force AI to find unreachable code, so I'm also cleaning out leftover AI slop.

140

Jon Harmon (he/him/his) @jonthegeek.com · 22/09/2026

I'm also finding that sticking to my standards helps the stuff that felt bad with LLMs feel better. At work, Claude & I brought test coverage in a #RStats 📦 that I hadn't worked in much before toward 100% and got rid of noise in tests, so now working there doesn't feel so much like slop.

140

Jon Harmon (he/him/his) @jonthegeek.com · 22/09/2026

I've had some helpful/hopeful replies in this thread (eg bsky.app/profile/hadl...). I also finally read this blog by @hadley.nz substack.com/home/post/p-... , and I wish he'd done part 1 of that talk one more time at #PositConf2026! It would've helped to offset some of the other AI. #RStats #dataBS

substack.com

y code when ai?

A talk on AI and coding

2171

Jon Harmon (he/him/his) @jonthegeek.com · 22/09/2026

I'm glad to see you feel that way! I'm definitely seeing a lot of judging the outcome over how the LLM gets to it. I'm not even sure that's a bad thing in many cases. If you have a way to judge accuracy, I don't know that programming has to require programmers anymore, let alone language specific.

110

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

I stopped reading @hadley.nz 's design substack religiously when it switched to all ai, but I just took another look and tidydesign.substack.com/p/y-code-whe... is extremely heartening.

tidydesign.substack.com

y code when ai?

A talk on AI and coding

010

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

I found it widespread, and I've seen plenty of evidence in code I've read. I don't know on an individual basis who has what personal practices. But "I did this thing before ai and that proves I'll keep doing that thing in this new world" is unconvincing. But...

110

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

I also said that the reaction when I mentioned to people that I still read all the code was "really? Why?"

100

Reposted by Jon Harmon (he/him/his)

-𝚖𝚘𝚡 @moxiest.art · 21/09/2026

remember when Biden replied to some right-wing bait question about how many genders there are with, "at least three. don't play games with me, kid." and then got elected president? that was their last winning strategy. dems: "you're a dipshit. fuck off" was your last winning strategy.

91368244

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

The more I think about it, the more crazy this makes me. Do you think you'll ever read another line of reshape code? I would be very surprised if anybody other than Claude or Kimi touches that code from now on. Other than misguided open source contributors.

210

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

GenAI has been as much of a step-change as the pandemic. What has happened for the last 21 years doesn't necessarily tell us what will happen next year. youtu.be/KkpKSqbGnBA?...

youtu.be

Max Kuhn - The Global Pandemic Ruined My Favorite Data Set

YouTube video by Lander Analytics

000

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

I honestly don't think it's a bad example for most cases. If you aren't working in a regulated environment, you SHOULD be vibe coding in rust. It's a cleaner, safer, faster language. But I don't know it yet and can't judge code in it, and I wouldn't ship it for science-adjacent things.

000

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

I think this is because I'd already leaned into air in Rstudio, so I didn't notice the difference when I finally moved (very recently)

210

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

Um. I don't know what indentation it doesn't have. It doesn't wrap roxygen docs nicely still, but I don't know what indentation thing it's lacking. I code in positron all the time now, and would murder someone if code didn't indent.

120

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

This would've been very confusing if you'd used em dashes here 🤣 But it's weird. I use agents sometimes to speed up dev. But I still read every line of code, and almost always end up fixing something myself. This got a lot of "really? Why?" in conversations at conf 😢

071

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

BTW: if bluesky existed during the industrial revolution, I'm 100% certain there'd be a nearly identical thread about weftcraft vs autolooms. But the autolooms haven't taken over yet, and there's still a place for quality, dammit!

3291

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

Yeah, it was fairly obvious they were doing the "we need a name python people won't run away from" thing, and I decided a few decades ago that "let's make a thing that could be used by the cool kids if only they didn't hate us!" is a bad way to position yourself.

151

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

Yeah, I work in a result-driven world enough of the time that I'd forgotten that a lot of people still want to be able to read and trust the actual code. And I'm very glad that's the case, cuz, I'm still working to make code that's cleaner and easier to understand. It's kinda my whole thing!

141

Reposted by Jon Harmon (he/him/his)

Greg Pak @gregpak.net · 20/09/2026

"My opponent wants to hire creeps to inspect your daughter's genitals if she wants to play soccer. I wouldn't trust him or any hired weirdo from his pedophile-protecting party alone in a room with any kid." Go on the offensive, you cowards! It's not that hard!

1347881232

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

I think most new code should probably be in typescript or rust, because they're popular and good languages that are well represented in training. BUT there's still a place for codecraft in science & and regulated envs! And wait, that's where I work now! There's a place for us #RStats weirdos!

8531

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

This feeling got way worse at #PositConf2026. PRs I opened at #PositConf2025 were still open. A keynote had "I don't write code anymore" as the (triumphant?) takeaway. The most famous R programmer had a talk about a thing he vibe coded in Rust.

10475

Jon Harmon (he/him/his) @jonthegeek.com · 21/09/2026

I've been feeling pretty down. When I found #RStats, I kinda fell in love, and decided to become as much of an expeRt as I could be. In the age of AI, that's not a good idea; why make the bots program in a language they don't really understand? #dataBS

7939

Jon Harmon (he/him/his) @jonthegeek.com · 17/09/2026

My #PositConf2026 was an accidental sequel to this 2019 @hadley.nz talk: www.youtube.com/watch?v=Qne8... He used the Design of Everyday Things metaphor, introduced {vctrs}, and admitted it might be too strict. That's why I made {stbl}. I hadn't seen this talk when I submitted my abstract! #RStats

youtube.com

[DSC 4.0] What Makes a Good Function - Hadley Wickham keynote

YouTube video by Data Science Conference

0140

Jon Harmon (he/him/his) @jonthegeek.com · 17/09/2026

BTW, the deck is definitely made to go with a live talk right now. I'll try to flush it out with more of what I said out loud in the next week or two.

000

Jon Harmon (he/him/his) @jonthegeek.com · 17/09/2026

I'll be stalking {stbl} a bit on CRAN and github, but please let me know if you use {stbl}, and if you find anything that could work more smoothly. And I guess it'd probably also be great to hear if you use it painlessly! #RStats #PositConf2026

061

Jon Harmon (he/him/his) @jonthegeek.com · 17/09/2026

Thank you so much for the kind words! I hope people are able to do cool things with {stbl}. There's a lot there that I couldn't fit into the talk! Of course now I'm scared that people will use it and I'll have done something stupid and be exposed as a fraud. Hurray impostor syndrome! 🙃

120

Jon Harmon (he/him/his) @jonthegeek.com · 17/09/2026

Thanks! That slot was torture. I didn't get to have the "yay, that's done!" moment until the conference was basically done!

020

Jon Harmon (he/him/his) @jonthegeek.com · 16/09/2026

Ahhh, no more looping through The Design of Everyday Things at 2x to prep for my talk! It's a good book, but it definitely getting to be a bit annoying 🙃 #PositConf2026

The audio book "the design of everyday things" in the audible player, set at 2x playback speed. Currently 39 minutes into the book, with 5 hours remaining at 2x.

120