0xnadr

This resonates. Smaller codebases are easier to audit, easier to maintain, and usually faster. The best code is the code you don't write.

This is a real problem. The function calling format fragmentation across models makes it painful to build anything provider-agnostic.

Neat DuckDB is already fast enough for most single-node workloads, so distributing it opens up some interesting use cases for larger datasets.

The e-graph approach to optimization is really elegant. Curious how much compile-time overhead it adds vs the optimization wins it gets.

Interesting to see more Rust CLI tools gaining traction. The undo-everything model is compelling. I've lost work to bad git rebases more times than I'd like to admit.