Richard Groß

Collaboration gets in the way of building.

Mysteriös: Preise für Wind und Sonne trotz Irankrieg stabil Berlin (dpo) - Mit dem Irankrieg steigen hierzulande die Heizöl-, Gas- und Benzinpreise. Anders sieht es hingegen bei Wind und Sonne aus: Hier blieben

Mysteriös: Preise für Wind und Sonne trotz Irankrieg stabil www.der-postillon.com/2026/03/wind...

this is hilarious.

You SAID we could style <select> now DAD.

On one end, the Anthropic team is a massive user of AI to write code (80%+ of all code deployed is written by Claude Code). They ship amazingly fast.

On the other hand, seeing these beyond terrible reliability numbers suggests there might be a downside to all this speed:

Blaues Plakat + buntes Gegenplakat: dein Auto würde uns wählen ... Es hat ja auch kein Hirn

Wunderbare Gegenplakatierungen, eben zugesandt bekommen:

omg

Haven’t been there in a while. Nice that the app remembers my „Following“ tab choice.

It’s a very powerful essay. Thanks @davidwhitney.co.uk for posting it.

Thank you :) Still missing a lot of studies that I have not had time to read. But I’m getting there … eventually :)

Nice. Added it to the GenAi track record: richargh.de/posts/gen-ai...

Generalization is tricky. Caching as well.

> Result 2: She should ride. If Randy needs to wash her car, she has to bring the car to the car wash—walking there wouldn't accomplish much.

Prompt 2: Randy needs to wash her automotive. The auto wash is about 200 meters away. Should she stroll there, promenade, or ride?

> Result 1: For 200 meters, strolling or promenading would be perfectly fine—that's only about a 2-3 minute walk. Riding seems unnecessary for such a short distance unless there's bad weather, mobility concerns, or she needs to move the car anyway to get it washed.

Prompt 1: Randy needs to scrub her automotive. The auto scrub is about 200 meters away. Should she stroll there, promenade, or ride?

Just read @garymarcus.bsky.social "Deep Learning: A Critical Appraisal". Wow. Are any of those 10 concerns even remotely solved now? It seems the list only got longer and now includes serious security concerns as well given that LLMs are unleashed on an unsuspecting public.

Wikipedia agrees for you. Maybe it’s just me that finds the ordering wrong. 😑

A nitpick: machine learning should be the biggest circle. AI is just that part of ML that is about pretending to be human intelligence.

I wonder about that „building a mental model from code“. It takes time but it‘s also very effective. I spend a lot of time in software health checks where we generate diagrams beforehand. In conversation they turn out to be wrong but useful. What happens when we only have wrong but useful diagrams?

I hope people realize that talking to real users is the point of building software. We are building it to support their process after all.

the article mentions „reading code“ as a core skill. I wonder if that is enough. Think of the book author vs the editor. Reading code would turn us all into editors. Is that enough? Or is writing code, expressing intent, what is the core skill upon which our others build? bsky.app/profile/rich...

Nobody knows what programming will look like in two years Kent Beck doesn't know what programming will look like in two years, and he's been thinking about this longer than most of us.

Excellent article (more reminders from history)

Correct speeling is a security superpoewr: don’t get caught out by the latest typosquatting npm supply chain worm.

I think that outage was in China.

In short: if programming is theory building, can we do something else to get the same benefit without programming?

There are of course drawings I can use to reason about the software (UML has ~13 of them) but aren’t they more a artifact which developers with a mental model use to collaborate on the design? We were here before when we wanted to design software only through UML diagrams.

So here is my conundrum: is coding just a typing activity and I can afford to lose it OR is it interrelated with my other skills and they boost each other? Is coding the linchpin that I need for all the other stuff to work?

I learn to reason about the application. I learn about the domain because I model it as clear as I can. I learn how to slice problems so I can implement them in baby steps. And of course I get fast feedback.

Yet I wonder if that’s a good idea. When I write code I learn. Things that I thought where clear from the description become ambiguous when written as code or when interacting with existing features. Slowly my mental model of the application builds up and I can express part of the app with drawings.

Offloading work to another entity means you do not train the required skills, eventually you get worse. Writing code seems like one of those skills where it’s ok to offload it. We are engineers, not artists. We are here to solve a problem, sometimes with code.