miniblog.

https://tigerbeetle.com/blog/2025-02-27-why-we-designed-tigerbeetles-docs-from-scratch/ has an interesting distinction between "physical" and "logical" hash of a tarball. By storing the hash of the decompressed tarball contents (i.e. the logical hash), they can verify the validity of files without needing to keep the tarball around.

Assertions are a surprisingly nuanced design space. In a test, if I assert `x < y`, I really want to see the values of x and y when it fails. Do you define an API for every possible predicate (Python's assertLess, expect.js) or try to support the native syntax (c.f. pytest)?

The word "agent" is so overloaded in the AI space. Sometimes it means a sophisticated interaction system, but other times it just means API. I think it's partly a sign of how new the space is. We don't have consensus on the best way to use these systems yet.