July 2, 2026
Ctrl-Alt-Delusion
The gauge broke: devs felt 20% faster with AI, measured 19% slower
Coders swore AI was a turbo boost — then the stopwatch served pure humiliation
TLDR: A study found seasoned developers using AI felt faster but actually finished work more slowly, raising fears that the tool gives a false sense of speed. In the comments, believers called the research outdated nonsense, while skeptics argued AI only helps for certain kinds of work — and the brawl got spicy fast.
A fresh study landed like a pie to the face of the “AI makes everything faster” crowd: experienced software developers felt about 20% faster using AI helpers, but when researchers actually timed them, they were roughly 19% slower. The article’s big warning is simple enough for anyone to get: the tool can make work feel quicker because text appears fast, but the real time gets eaten up checking, fixing, and second-guessing what the bot spits out. In other words, the speedometer may be lying.
And the comments? Absolute food fight. One camp basically yelled, “nice try, grandpa”. User jiggawatts mocked the piece as being from the “stone ages,” bragging they used a cutting-edge model to do six months of work in two weeks. Another commenter flat-out declared the debate over, insisting AI productivity is “no longer up for debate” and anyone still arguing is just behind. That’s the swagger side.
But the skeptics came armed too. One person side-eyed the whole study with the devastatingly short jab, “16 developers across 246 tasks” — a tiny-sample dunk that says a lot with very little. Others argued the whole fight is messier than a single number: AI may be amazing for repetitive busywork, but much worse when the job is really about understanding a giant existing project. There was even a fun throwback comparison to old studies where people using keyboards felt faster than mouse users while actually being slower. So yes: the numbers started the argument, but the real show is the comment section split between AI evangelists, nitpickers, and weary realists all insisting they’re the only sane ones in the room.
Key Points
- •The article says a METR randomized controlled trial found experienced developers felt about 20% faster with AI tools while measured performance was about 19% slower.
- •The trial described in the article involved 16 experienced open-source developers completing 246 tasks in codebases they already knew.
- •The article states that the cited study includes caveats, including that results may differ for junior developers and for greenfield work.
- •The article cites Faros AI, DORA, and GitClear data as showing increased code generation, review burden, churn, and delivery instability with little net delivery improvement.
- •The article argues that AI lowers the cost of code generation but shifts the bottleneck to verification and review in existing codebases.