The case for more ambitious software

Original: We should be more ambitious with the software we build

Why This Matters

AI tooling is quietly raising the ceiling for what solo devs can ship.

Developer Dan Luu argues that AI-assisted tooling now makes once-infeasible rewrites practical. He cites pgrust—a Postgres reimplementation—which has already surfaced real Postgres bugs during testing.

Dan Luu's essay, prompted by conversations with Michael Malis (creator of pgrust, a Postgres reimplementation), argues that developers should aim higher now that AI coding tools have shifted what's feasible for small teams or solo builders.

The piece uses pgrust as its central example—not because it's polished, but because Malis's testing discipline has been thorough enough to expose genuine bugs in production Postgres: a numerical bug breaking valid SP-GiST operations, a format cache bug causing incorrect JSONPath results, a pgcrypto failure bug, multiple memory-safety bugs in the dictionary loader, and a money overflow bug. Even if pgrust never ships widely, those upstream fixes are durable improvements.

Luu acknowledges the reflexive skepticism around rewrite projects is justified—most are, in his words, 'vibeslop garbage.' The distinction here is rigorous testing and benchmarking, not the ambition of the project itself.

For his own part, Luu mentions building a custom regex engine tuned to outperform general-purpose libraries on specific workloads, and a ripgrep fork, after realizing that faster AI inference would expose bottlenecks in syscall overhead, compilation, and file search. His framing: the question has shifted from 'can a small team do this?' to 'will they test it properly?'

Source

danluu.com — Read original →