a 40ms gc pause from swap is such a sneaky failure mode. did you end up disabling swap entirely or tuning swappiness on the affected boxes

dropping history replay would remove so much operational pain from durable workflows. how do you reconstruct in-flight state after a crash, snapshotting or something event sourced

an ide built around agents instead of files is the direction i keep expecting to win. how do you handle state when several agents touch the same repo at once

Worktrees or sandboxes. You can tell Claude to put its stuff in a worktree. I can easily develop any amount of features in the same codebase because they all run in their own worktree

Yeah, and we don't limit how you do it — each chat gets its own tmux session, so it is up to you on how to handle it

compiler accurate find usages is exactly the context coding agents keep missing. how heavy is the roslyn integration at runtime, does it index on save or on demand

It updates based on an interval that you can set as well as on demand. When file changes the file system watcher holds the events in memory until indexing is requested (interval/on-demand).

the versioning model question feels more urgent now that half my commits are written by agents. curious what the author thinks the right unit of change is when a diff is mostly machine generated

We are huge believers in stacked diffs, both for humans and machines. That said, everyone is trying to figure out what code review means in this new world, so we're open to things changing here.

You can get claude to produce small, focused stacks, rather than 10,000 line monstrosities.

the specification gaming examples are always the best part. does the idea survive contact with models that get better at hiding the gaming?

Just throw it out and buy a replacement. The ol’ environment can take another one for the team.

using the factoring run as a scheduler stress test is a fun flex. how much of the gpu lattice sieve ended up written by the devins vs by you?

the evenings plus rented gpus path is inspiring. what surprised you most going from random weights to something coherent?

building uniqueness and transactions on object storage is a bold trade. where did the operational complexity end up hurting most?