Rust + Tauri as an agent cockpit: what I'd do differently

Agents that write into real systems need a place where humans watch them work. We built ours as a Tauri 2.0 desktop app, Rust core, React/TypeScript front end, running on operators' Windows machines. A year in, verdict: right call, three scars.

Right call because: the binary is small, auto-update via Tauri's updater has been boring (the highest compliment), and Rust on the desktop side means file handling, watchers, and OS integration don't have a runtime to babysit. Operators treat it like a native tool, because it is one.

Scar one: IPC boundaries. We started with chatty frontend↔core messaging and paid in serialization overhead and race conditions. Do fewer, coarser commands; treat the Rust side as a server with a real API, not a bag of helpers. Scar two: state. The desktop app, the backend, and the agent each believed something about the world; reconciling three beliefs is a distributed-systems problem you have opted into. Pick one source of truth early and make the other two caches. Scar three: Windows is the environment, not a target. Test on the real machines, locked-down enterprise Windows with antivirus and a proxy, from week one.

Would I use Tauri again? Yes, but I'd design the IPC contract before writing either side, the same way I'd design an API between services. Because that's what it is.