Shared test environments vs per-test isolation — how do you decide?

We have two camps on our team: **Camp A (shared environment):** Tests run against a shared staging server. Fast to set up, tests reflect real infrastructure. Problem: tests interfere with each other, ordering matters, data gets corrupted. **Camp B (isolated per test):** Each test (or test module) gets its own clean state. Reliable but: Docker compose takes 30s to start, running 400 tests takes forever. Is there a middle ground? I'm thinking: isolated DB per test worker, shared application servers. Does that actually work or does the app server become the shared state problem instead?