We should be more ambitious with the software we build

What the article says

The author has watched Michael Malis work on pgrust, a project that rebuilds Postgres, a popular database. Many people dismiss projects like this as low-quality AI-made code. The author says he trusts Malis, and he checked the project's work. He found that Malis's testing is thorough enough to find real bugs in Postgres itself. The author's main point is that people who build software should be more ambitious, because new tools make big projects possible.

Key facts

• pgrust is mainly pitched as a faster version of Postgres.
• Testing of pgrust found several real bugs in Postgres, including a numerical bug, a format cache bug, a pgcrypto bug, memory safety bugs, and a money overflow bug.
• The author imagines AI tools that run 10x to 100x faster, and lists slow spots such as ripgrep, Rust compiling, and syscall and filesystem overhead.

Why people doubt it

Many new AI-made projects look impressive but do not work well. Many also claim speed using weak tests. So it is hard to tell which projects are real. The author says a project can still be good quality if its creator tests and measures performance carefully.

Why it matters

The story shows that AI tools may let small teams or single builders attempt projects that were once too big. Even a failed attempt can help the wider software world, because the bugs found along the way can be fixed in tools people already use.

What to watch

• Whether pgrust reaches a useful, reliable version.
• Whether its speed ideas move into Postgres or other databases such as Clickhouse.
• Whether more builders test their AI-made software carefully enough to earn trust.