We should be more ambitious with the software we build

The central argument is that advances in AI make it practical to attempt software projects that programmers have long avoided as too costly, and builders should raise their ambitions accordingly. The example of pgrust, a Rust rewrite of PostgreSQL, shows why ambitious experiments can matter even if they never become successful products: careful testing has exposed real PostgreSQL bugs, while its performance work may also yield useful optimizations. Distrust of low-quality AI-generated projects is justified, but dismissing every rewrite as “slop” overlooks what disciplined testing and serious engineering can accomplish.

The author describes building a specialized regex engine, a faster ripgrep fork, and a Rust interpreter to improve personal development workflows - projects that once might have taken a year but now took only hours of personal effort with AI assistance. Similar gains appear in game-playing AI and in models implementing complex systems such as zstd, IMAP, and Pandoc. Yet more capable models do not make software effortless: the author finds that they enable more ambitious work, while checking whether implementations are truly correct remains difficult. The broader obstacle is an outdated sense of what is worth building; tasks once dismissed as impractical can now be attempted, though turning a personal tool into a reliable product still requires substantial testing and refinement. The performance comparisons described are mostly informal, not comprehensive benchmarks.