Dan Luu
An engineer and writer known for carefully re-testing widely repeated claims about software, and often finding them wrong.
Dan Luu writes a long-running blog that does something rare: it takes claims everyone repeats and actually measures them. His background is in hardware verification and performance work, and that shows in the method. He states his guesses before looking at results, publishes the flaws in his own tests, and refuses to draw conclusions bigger than the data supports.
In 2026 he turned that approach on AI coding claims, including the popular advice that compact dynamic languages are cheaper for coding agents to work in. Running larger, more realistic tasks, the effect largely disappeared, and one of the benchmarks people cite turned out to have a broken test setup that made a whole language look bad. His writing is a useful antidote when AI benchmark charts start circulating.