Agent Workload Design

Agent workloads are not simple sequential reads:

1. Agents are not sequential readers. They repeatedly perform small reads, partial writes, directory scans, metadata lookups, and artifact updates.
2. Agents constantly scan project structure. Traversal, stat, open, read, write, rename.
3. Agents produce many intermediate files. Plans, logs, patches, tool outputs, reports, caches, indexes.
4. Agents share state. One agent's output is another agent's input — the filesystem becomes the coordination layer.
5. More agents amplify random IO. Concurrency multiplies metadata pressure and workspace updates.

BrewFS's design answers: a shared POSIX-like namespace, a RandRW-optimized data path, and metadata backends selected for concurrent access.