Underrated agentic workflow: take a messy commit history and slowly rewrite it into atomic revisions that can be reviewed and merged independently.
Works beautifully with jujutsu (jj), using a duplicated branch to ensure no destructive changes are made to the original PoC.
Little update on my reflection adventures: I have extracted the core components of Pavex's compile-time reflection engine as a set of standalone crates!
It serves as a preliminary step ahead of another release in the coming weeks 🛠️
That weird urge to go build a new CI platform.
Pure control plane for task orchestration, plug-your-own-compute with a marketplace of vendors, easy to bridge to GHA for initial adoption.
But it also sucks to perform commit surgery on a branch with 80+ revisions, which you intentionally created to be able to split this into granular PRs which can be sanely reviewed.
With a bit of AI support, you can improve the life of the reviewer with minimal overhead.
New year, new vacancies: Mainmatter is looking (again!) for an experienced Rust developer to join our Rust migration projects.
C, C++, Delphi. Different starting points, same target: Rust!
If you're interested, send an email to the address in the screenshot.
The mainstream conversation around AI agents is stuck on "AI slop."
There's more to it: we can use cheap agent code to do quality-raising work that didn't make economic sense before.
What's viable now that writing code isn't the primary bottleneck?
lpalmieri.com/posts/agenti...
Latest occurrence: I'm currently working on a trait redesign which requires touching _every single impl_ in the project (>30).
As you work through the impl, you realize some tweaks are needed to the design and start accumulating clean-up commits. It'd suck to review that.