- Author
- matklad
- Published
- May 12, 2026
- Kind
- Personal blog post, a reply to an emailed question
- Original
- matklad.github.io/2026/05/12/software-architecture.html
- Subject
- Software design
What it is
The author, who writes as matklad, replies to a physicist who works on scientific code. The post gives two meta observations and then concrete recommendations.
What it says
- The post says software design is best learned by doing, not by courses.
- Conway's law matters, the author says, since software repeats the organization's social structure.
- The author suspects scientific code differs from industrial code because of incentives.
- For rust-analyzer, the author isolated features behind catch_unwind and kept the core stricter.
- The author knows no single book with the truths and lists talks and blogs instead.
What evidence it gives
The post offers anecdote and opinion. The main example is the author's work on rust-analyzer and IntelliJ Rust. The post quotes another engineer, neugierig, on social issues. It has no data or tests.
Interests and limits
This is one engineer's view, given to one asker. The author led the projects used as examples. The post admits plans changed in ways that were inconvenient. It does not test whether the advice works for others.