Post #2177552

2026-03-28 12:12 UTC

I've seen this now I'm pretty much all of Redhat's job Postings: What you will bring A willingness to proactively utilize AI-assisted development tools such as Cursor and Claude Code for code generation, auto-completion, and intelligent suggestions to accelerate development cycles and enhance code quality.Open to exploring and experimenting with emerging AI technologies relevant to software development, proactively identifying opportunities to incorporate new AI capabilities into existing workflows and toolingand i have a serious question: is there any research showing irrefutable proof that using these tools does at least a good a job at "accelerating development cycles", "enhancing code quality" as "legacy methods" such as "code review" or "pair programming" do? these methods, which, in addition to teaching one or two people something about the code base they're working on, also provide opportunity for creating team cohesion (or conflict, but at least there's some communication!) so, how do LLM tools stack up against that? what other advantages do they bring to the table, if not these? in medical research, you're supposed to show that you're new medication is better than the current established best treatment. some pharma companies try to get away by showing that their new drug is better than a placebo (nothing), and i have a feeling that's exactly what's happening with AI tools. or am i just very biased?

Replies (3)

@pmdj@mstdn.social 2026-03-28 12:26

@meena@cathode.church In medical research, you first have to show a treatment does no harm. (Or at least exhaustively enumerate and quantify the risks, then decide whether it’s ethical to proceed with proving efficacy.) I have my doubts the chatbots could clear that hurdle if such a standard were applied.

Open ##3031710
@mia@tearoom.tearmoon.com 2026-03-28 12:32

@meena@cathode.church there is research that shows that degenerative BS usage does not result in statistically significant improvement to a software developer's performance (so when the developers themselves say they are 10%, 30%, 1000% more effective they are being extremely subjective and deluding themselves) the same research (mostly done by the very companies that peddle this stuff) shows that more often than not it results in performance degradation

Open ##3031711
@ohno@ap.dny.social 2026-03-28 13:58

@meena@cathode.church There is a mounting body of evidence that shows that: Using AI makes you lose skills you have and makes you dumber in general. The latest one I’m aware of https://arxiv.org/abs/2506.08872.Using AI stops you from developing new skills, which this article gets into based on some recent studies https://www.psychologytoday.com/us/blog/the-algorithmic-mind/202603/adults-lose-skills-to-ai-children-never-build-themWhatever productivity gains top performers might get out of it is eclipsed by the increasing amount of BS everyone else can now generate and throw at them at light speed. So good luck to Red Hat!

Open ##3031714