François
Thomas 🔭🕹️That study that shows that developers think they save 20% of time when using coding LLMs, but they’re actually taking 20% longer is so funny.
Not only does the AI bullshit the developers—the developers also bullshit themselves.
Liminal witch 🧙♀️ Sarah@thomasfuchs is there a paper/article with sources I could read about that stats? I’m not disputing that number but there’s been too many magic numbers thrown out about that stuff
Florian 'floe' Echtler
Jon Lasser is Hanging in there@thomasfuchs @lisamelton I imagine in part it’s like CLI vs GUI, where CLI takes longer by the clock but the human brain often experiences it as faster because of the kinds of processing it’s doing. If you have a tight dev/test loop, it probably feels fast.
Pēteris Krišjānis@thomasfuchs but that is a thing. Coding is incredibly hard to measure objectively. And it drives production nuts, this is why they obsess over genAI scam promises so much.
In reality, there is no other way than actually teach your devs to be efficient through specific tooling, templating, review process.
It takes time and money, but it makes your coders much faster and better.
Bean counters are incredibly frustrated that they can't throw money at this problem to make people disappear.
In reality, there is no other way than actually teach your devs to be efficient through specific tooling, templating, review process.
It takes time and money, but it makes your coders much faster and better.
Bean counters are incredibly frustrated that they can't throw money at this problem to make people disappear.
billy joe bowers-8647This is interesting, because I just had a conversation with someone about using AI for little chores and one of the things I mentioned was "Are you sure you're really saving time?".
I told them about what I call "computer time", the time people don't count when they're working on things.
The example is smart phones; many times I had people saying "Oh, I've got that right here!" and then spend literally 20 minutes trying to find something.
"Hold on a second..."
brood of byrdes@thomasfuchs
"We conduct a randomized controlled trial (RCT) to understand how early-2025 AI tools affect the productivity of experienced open-source developers working on their own repositories. Surprisingly, we find that when developers use AI tools, they take 19% longer than without—AI makes them slower."
https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
Ur Ya'ar
Metalpoet 
@thomasfuchs twice this week I had test runs fail far from the changes I was testing. On both occasions it turns out that the the copilot I didn't ask for which was auto installed on my work system had added a line I hadn't noticed - in both cases it was a line that did need to.be there, only copilot hadn't seen that I knew this and had already added it.
In a pipeline, where duplicating a structural statement is a syntax error