Talk Is Cheap: The Operational Impact of LLM Use

A spicy new write-up just dropped a bucket of cold water on the “AI makes everyone 10x better” fantasy, and the comments immediately turned into a full-on workplace group chat meltdown. The article points to data from Faros, a company that tracks how software teams actually work, covering 22,000 developers across 4,000 teams. The headline finding? Individual workers using large language models — the chatbots behind today’s AI tools — seem to finish more tasks. But the bigger picture is uglier: teams ship updates less often, work takes much longer to reach customers, and quality appears to get worse. In plain English: people may feel faster, while the company gets slower.

And oh, the commenters had thoughts. One camp basically yelled, “Finally, receipts!” with one person cackling at the author’s line, “It’s hard for me to put into words how bad this is,” and calling it the evidence “reasonable people” were waiting for. Another group pushed back hard, saying this is just AI’s awkward teenage phase. As one commenter put it, the industry hasn’t even reached “junior level” in knowing how to use these tools properly yet. Others said the pattern isn’t even unique to AI at all — it looks a lot like what happens when you throw more people, contractors, or chaos at a late project and call it a strategy.

The funniest mini-drama came from a brutally simple question: if your employer stopped paying for AI tomorrow, would you actually need it enough to pay out of pocket? That one landed like a grenade. Suddenly the debate wasn’t just about hype — it was about dependency, denial, and whether the shiny assistant is a lifesaver or just making a bigger mess faster.