> When developers say that LLMs make them more productive, you need to keep in mind that this is what they’re automating: dysfunction, tampering as a design strategy, superstition-driven coding, and software whose quality genuinely doesn’t matter, all in an environment where rigour is completely absent.
https://www.baldurbjarnason.com/2026/the-two-worlds-of-programming/
by @baldur
The stock price of the company that caused worldwide outages and economic havoc, Crowdstrike, even in a stock market affected by the Iran war, is higher today than its peak before the outage.
https://www.baldurbjarnason.com/2026/the-two-worlds-of-programming/
Heh, just as I expected, I just forgot to keep updated on that.
The other thing I “prophesised” is that in 2–3 years since the Crowdstrike incident there will be another similar fail – we'll see, I'd love to be wrong, but in this state of IT, it seems I won't be 🙃
@davidgerard "Superstition-driven coding in an environment where rigour is completely absent" — this is the uncomfortable truth. The productivity gains are real but they're measured in lines shipped, not in production stability.
The question nobody asks: are we 10x more productive, or are we creating 10x more things that will break?
@samstart @davidgerard GenAI makes it so easy to generate LOCs that its useless as a measure (it was useless before, it's even more useless now)
If you think about the tech debt that is racked up by devs without AI in the current programming environment and you make that environment 10x faster, you should expect that your tech debt is at least 10x what it was.
(It may not be one-to-one either - I wouldn't be surprised to see tech debt increased 100 fold by AI outputs)
@jer @davidgerard The 100x tech debt multiplier is an interesting thought. If AI generates 10x the code but skips the structural work (tests, CI, dependency management), the debt isn't just 10x — it compounds because each untested module interacts with other untested modules.
The failure modes multiply faster than the code volume. That's why catching structural gaps early matters more, not less, as velocity increases.
@davidgerard @baldur As we’re once again riding the ’Gartner Hype Cycle’ it becomes abundantly clear that there is indeed a binary among software developers. Those of us who consider the craft itself important, and those who simply don’t.
What makes this iteration dangerous is that it’s practically only later group who are onboard with it, while the former are unanimously (and with a growing desperation) reaching to for the emergency breaks.
@davidgerard @baldur can someone please think of the poor cloud providers? Even if the producers of source code who use LLMs blindly slouch towards the apocalypse of bloated buggy code everywhere, they represent a fine market to the cloud providers willing to provide a few more instances and databases to handle the load.
Here speaks a man who trimmed 60% or more from a client’s database spend by adding indexes to their database. I suspect LLM generated code / schemas would be better than what I got handed, but that is a low bar!
David Gerard
LEdoian
Marco Hengstenberg
patter
DeManiak 🇿🇦
sam
Jer
alanxoc3
Bolet modo Sombrero de paja 🏴☠️ 
🌸 Vårgubben
MikeStok