Baldur Bjarnason

The two worlds of programming: why developers who make the same observations about LLMs come to opposite conclusions: https://www.baldurbjarnason.com/2026/the-two-worlds-of-programming/

This time of year, when the ravens congregate in town before scattering into the countryside for the summer, I tend to stalk their regular hangout spots. #raven #ravens #iceland #nature

This raven is picking the meat off a sheep's jaw. I'm guessing somebody threw their "svið" leftovers out for the ravens to enjoy. #birds #iceland #raven

So, for this #caturday, the reason why I'm thankful my neighbour has placed his bird feeders well out of the reach of the cats.

“"AI layoffs" like Block's are a race to the bottom that have nothing to do with productivity.”

https://productpicnic.beehiiv.com/p/ai-layoffs-like-block-s-are-a-race-to-the-bottom-that-have-nothing-to-do-with-productivity

> But after those layoffs come waves of rehiring

The only thing I'd add to this excellent post is that rehiring will only happen if the economy is otherwise in good shape, which is by no means certain at this point. (Oil crisis. Disasters. Trade wars. Etc. Etc.)

RE: @glyph@mastodon.social

I’ve never used any of the LLM chatbots, avoided them even in the early days, because they had all the hallmarks of a major cognitive and psychological hazard from the very beginning.

“I’m not gonna use this shit until I can be sure it won’t mess me or others up” is an entirely reasonable position to take, but for some reason it makes even those who are otherwise lukewarm on the tech quite angry.

The LLMs in coding debate over the past few years:

“They’re taping shotguns to their feet and attempting to use them as stilts! That’s a recipe for disaster”

“IDK. My friends say they’re the perfect length to get the height they need. They say it’s a good idea and my friends are smart.”

“Amazon holds engineering meeting following AI-related outages”

https://www.ft.com/content/7cab4ec7-4712-4137-b602-119a44f771de

Magnify dysfunction through automation and all you get is more dysfunction.

“The first AI agent worm is months away, if that – Dustycloud Brainstorms”

https://dustycloud.org/blog/the-first-ai-agent-worm-is-months-away-if-that/

Because there is a high chance that once this happens, it’s going to backdoor itself into many other systems that didn’t opt in to AI agents.

Seeing all the “Björk is overrated” reactions to Rosalia’s Brit Awards performance takes me back to circa 2000 when I had just moved to the UK. Back then, when people found out I’m Icelandic the most common reaction I got was an immediate “I hate Björk”

Second most common reaction?

“I love Björk”

My posts on Bluesky seem to have caught the attention of the vibe coder crowd.

Current highlights:

- You aren’t allowed to worry about HarfBuzz unless you’ve written something like it that’s as popular
- linking to a mastodon thread is abusive (fair, I’ll give them that one
- You should be grateful
- You don’t know what you’re talking about

Mostly, they seem to be an assortment of comments in the ‘all criticism is abusive’ vein. 🤷🏻‍♂️

If you had any doubt that the rise of LLM tools is a threat to F/OSS, even beyond the fact that its trained on it without permission, and is now frequently used to replace it (why import a battle-tested library when you can have an "agent" half-ass it?), people are now using LLMs to create derivative rewrites of open source projects to give them cover for bullshit relicensing attempts.

https://github.com/chardet/chardet/issues/327

“Propellant. — Ethan Marcotte”

https://ethanmarcotte.com/wrote/propellant/

> Because what all this corporate flexibility means is that — in addition to all their other many failings — these “AI” platforms are machines of war and death. Anyone infuriated by the military-industrial complex should stop supporting them.

“Artisanal care”

https://tante.cc/2026/03/04/artisanal-care/

> Software was doing bad before, standards of quality being largely nonexistent. But “AI” and the promise that you can just magically create software is pouring gasoline on the fire

(Following thread was prompted by people pointing out that the Bluesky dev team seems heavily into vibe-coding now and originally posted on said vibe-coded Bluesky platform that is now constantly failing.)

Over the past year, every single time one of the apps or services I use suddenly became less reliable and more buggy, I never have to look far for the "Claude is amazing and now writes most of my code" post for the devs involved.

"Ars Technica Fires Reporter After AI Controversy Involving Fabricated Quotes"

https://futurism.com/artificial-intelligence/ars-technica-fires-reporter-ai-quotes

Seeing a lot of people assume that this reporter got fired for making one mistake during a sick day, something most people could see themselves doing

I think it's safe to assume that the reporter's version of events downplayed his culpability to a substantial degree

“How to grow strawberries • Buttondown”

https://buttondown.com/monteiro/archive/how-to-grow-strawberries/

> Sadly, I fear those things are close to impossible to find in the tech industry at the moment.

The problem with looking at global trends instead of local when it comes to employment is that things are always unevenly distributed. When manufacturing jobs in the US collapsed, the fact that manufacturing jobs grew worldwide didn't help US workers.

If dev hiring ever recovers, which is going to be likelier?

1. companies will rebuild their domestic dev capabilities.

2. They offshore everything.

Sometimes all you can do is go out and capture the snow #photos #iceland #nature

“Five Things: February 27, 2026 | As in guillotine...”

https://loudpoet.com/2026/02/27/five-things-february-27-2026/

All five links are interesting reads.

“Secret Agent Man”

https://2ndbreakfast.audreywatters.com/secret-agent-man/

> Some weeks, the education technology news is incredibly grim, and sorry to say this was one of those weeks. (Warning: this is a long email.)

"Objects not data: a photography and illustration print experiment"

https://www.baldurbjarnason.com/2026/objects-not-data-art-prints/

> Over the next few months, I’ll be experimenting with selling art and photography prints as a way to fund my work on this newsletter

Objects not data: a photography and illustration print experiment: https://www.baldurbjarnason.com/2026/objects-not-data-art-prints/

Current worry: even if I did believe that coding “agents” could be used productively and safely by organisations with robust engineering practices, neither WebKit nor Firefox are that and WebKit specifically has a knack for bad regression bugs in existing features

So, y’know, that they seem to be using these tools is a worry

To throw a hypothetical example out there: a coding “agent” that generates working code half the time and 10% of all the code (working or not) is actively insecure will obviously be trash to most who test it

But an “agent” that generates working code 95% of the time while still having 10% odds of the code being insecure is a security catastrophe—a nightmare, not a productivity boon

It’s not a revolution but a crisis in the making

“Are AI-generated summaries suitable for studying and research?”

https://www.tue.nl/en/our-university/library/library-news/24-02-2026-are-ai-generated-summaries-suitable-for-studying-and-research

But it’s already safe to conclude that one popular use case is, in fact, a bad one: AI-generated summaries.

This goes into some of the specifics why LLMs are in fact very bad at summarising

RE: @xgranade@wandering.shop

I’ve had basically an identical encounter to the one described in this thread a few times as well, both offline and online

“Rely On AI And Get Left Behind – Codemanship’s Blog”

https://codemanship.wordpress.com/2026/02/21/is-comprehension-debt-in-your-risk-register/

These are effects we’re seeing today. Most dev teams are shipping less reliable software, and shipping it later thanks to “AI”. Good work, everyone!

Anybody who thinks “quickly churns out high performing code!” and “mass-manufactures vulnerabilities and disaster UX designs” are mutually exclusive has clearly never met a programmer in their entire life.

Getting a bit depressed as, no matter how I slice it, basic economics would seem to indicate that the software developer job market is largely fucked no matter what happens now. No matter whether the LLM tools work or not and no matter what happens to the bubble

(short thread)

RE: @iris_meredith@mastodon.social

> Tech people, by contrast, are observably swayed extremely easily by almost any messaging that's presented to them.

This is a near-universal truth of the tech industry. Most tech people are worryingly malleable by authority in my experience and the theory in the quoted post would explain a lot

“As AI enters the operating room, reports arise of botched surgeries and misidentified body parts”

One FDA report in June 2025 alleged that AI software used for prenatal ultrasounds was misidentifying fetal body parts.

https://www.reuters.com/investigations/ai-enters-operating-room-reports-arise-botched-surgeries-misidentified-body-2026-02-09/

The push for the adoption of “AI” in healthcare and education is a push to dismantle healthcare and education for the masses.

It’s entirely possible that through oligopoly dominance, deregulation, and LLM capture of the dev ecosystem, genuinely working software will eventually be the exclusive domain of expensive specialised high-end industry-specific software and the occasional eccentric FOSS app maintained by a weirdo only known by a handle like ‘mojojojo55 ‘

The duality of every vibe-coded project you see and every LLM-generated pull request you review being dangerous insecure messes and then turning around and hearing every respected figure in the web dev community saying these tools are now better than human developers

If I hadn’t live through several brain-cooking bubbles (Iceland has had a few more than the rest) I’d be shocked at the dissonance

“Coding Is When We’re Least Productive – Codemanship’s Blog”

https://codemanship.wordpress.com/2026/01/30/coding-is-when-were-least-productive/

But if I hadn’t made that trip and seen for myself, and had a chance to talk to a department manager in that store, those three lines would have been applying special offers wrong

“If users notice your software, you’re already a loser – Pivot to AI”

https://pivot-to-ai.com/2026/01/10/if-users-notice-your-software-youre-already-a-loser/

If you’re adding sparkly icon “now with AI!” features to your software, you’re in quarterly driven failure mode

With every report about cities or states in the US pushing back against the violent authoritarianism of their federal systems, I’m reminded of Tainter description of the prelude to collapse in “The Collapse of Complex Societies”

There is, first and foremost, a breakdown of authority and central control. Prior to collapse, revolts and provincial breakaways signal the weakening of the center. Revenues to the government often decline. Foreign challengers become increasingly successful.

“Voters Were Right About the Economy. The Data Was Wrong.”

https://www.politico.com/news/magazine/2025/02/11/democrats-tricked-strong-economy-00203464

In other words, nearly one of every four workers is functionally unemployed in America today — hardly something to celebrate.

and

Our alternative indicator reveals that, since 2001, the cost of living for Americans with modest incomes has risen 35 percent faster than the CPI.

“Guest Gist: 2026, Our Already Rotting Future”

https://www.thegist.ie/guest-gist-2026-our-already-rotting-future/

The difference is I’ve not yet encountered a calculator which asks its users, every single time it is switched on: “how about 43 x 756 today?”, or “want to know the square root of 7,568?”. The idea would be preposterous, even if we could guarantee the numbers would be correct.

“A Deep Dive Into SVG Path Commands” https://www.nan.fyi/svg-paths

“Deepfake Porn Victims Are Seeking Federal Protections Through Legislation | Teen Vogue”

I honestly think that this is a much more important issue than the fictional problem of a language model going all SkyNet on us. https://www.teenvogue.com/story/deepfake-porn-victims-are-seeking-federal-protections-through-a-new-bill