Dan Luu is an engineer and technical writer whose experiments examine performance, testing and software-industry assumptions. His work connects hardware experience with practical questions about the behavior of computers and coding agents.[1][2]
Start with Do Dynamic Languages Really Save Coding Agent Tokens? →
Education
- Education
- B.S.; majored in Computer Engineering and Mathematics[2]
Links
Technical writing Mastodon Bluesky
X · danluu 
GitHub · danluu Career
Luu majored in computer engineering and mathematics and earned a B.S. before leaving graduate study. His career includes Micron flash-memory and IBM POWER6 internships, Centaur hardware work and Recurse Center in spring 2013. A 2016 QCon biography places him on Microsoft’s BitFunnel project after work on Google’s TPU.[2][5]
Contributions
Measure cost and correctness together
His language comparison asks agents to implement a zstd decoder from its specification without internet access, then evaluates them with hidden tests. He compares correctness with cost across languages and reasoning settings. An apparent dynamic-language advantage at medium effort becomes mixed at higher effort. Small programs that are cheap to print therefore provide limited evidence about larger engineering tasks. The article also inspects harness bugs that can make one implementation’s score appear to belong to another.[7]
Measure the response a person experiences
For Computer latency: 1977–2017, Luu used a high-speed camera to time a keypress becoming a character on a terminal. Some older machines responded faster than much newer ones. The comparison directs attention to the entire input, operating-system and display path, rather than treating CPU speed as a substitute for responsiveness. His tables and measurement notes let readers inspect the differences among the tested configurations.[8]
Tools, writing & talks
Writing, research & talks
Start here
Experiments comparing coding-agent cost and correctness on specification-driven tasks.
Start with the zstd experiment and compare the medium- and higher-effort results. Read the evaluation setup before generalizing a language ranking to your own tasks; basic testing knowledge helps.
Author[7]Original source- Computer latency: 1977–2017danluu.com
High-speed-camera measurements of end-to-end input latency across computers.
Author[8] - How I learned to programdanluu.com
A first-person account of internships, hardware verification and learning software development.
Author[2]