TL;DR: If a project used to be too expensive to attempt, it may be worth reconsidering the project—not just automating the old workflow.

In an October 24 essay, Dan Luu describes building personal tools he would previously have expected to take far longer: a regex engine, changes to search tooling and experiments with Rust execution. He credits Michael Malis for the provocation:

we should be more ambitious with the software we build

The useful tension comes later. Luu distinguishes something that works for him from a maintained product and says stronger models can leave him with more work he cannot easily verify. His appendix includes Yossi Kreinin's counterargument: people can become responsible for giant systems they cannot reason about.

A good discussion starter for anyone making things with AI: which project has become feasible for you, and what would convince you that it actually works?