Dan Luu: AI made a year of work take hours, so build something harder

The evidence Luu offers

In his post, Dan Luu quotes Michael Malis: "we should be more ambitious with the software we build." Most AI-written projects are low quality, he concedes, but careful testing can make the output good enough to use.

His examples: Malis's pgrust, a Postgres rewrite that Malis says is not ready for production but which has found real Postgres bugs (a numerical SP-GiST bug, a JSONPath format cache bug, a pgcrypto bug, memory-safety bugs in the dictionary loader and a money overflow bug). Luu's own regex engine tuned for his workloads, a ripgrep fork using its compiler and a Rust build speedup. A game AI for Azul that reached superhuman play in about 5 hours. A zstd implementation that passed over 99% of registered tests for roughly $1.

He is also skeptical of 100x productivity claims, citing Amdahl's law, and notes Yossi Kreinin's counterpoint that most people should be less ambitious because AI-written code is often overfit and hard to verify.

The thread: bored, or just unhappy?

"Much easier to complain that you are bored than to step back up to high-effort with your new fancy tools." — mindcandy

The sharpest rebuttal was about the tribalism, not the tooling.

"It makes me feel actually sick that so many people can't separate their positive feelings about AI from people's misery." — happytoexplain

Jacques Mattheij, who builds a 32-bit protected-mode OS predicted fewer people will share their projects if the end result is just feeding the machine. And one commenter simply described the loss.

"It most definitely took me probably ~2x longer than it would with an LLM, but it was much more enjoyable owning every decision behind the process" — xpct, an anonymous programmer who still hand-codes for enjoyment

Projects with zero business case

The best replies were builds. One commenter said he could not have found the time to remake a 1995 4X game he loved as a kid, and another wrote a JavaScript engine from scratch as a product developer who knew only basic parsing. Luu's test is whether your last AI-assisted project would have been rejected by your own estimate of the time it needed.