Is there a practical upper limit on how long a single automated test should run?

We have a handful of E2E tests that take 4–6 minutes each. They cover complex user journeys (onboarding flow, subscription checkout, multi-step report generation). These tests are valuable — they've caught real bugs. But they're also: - Too slow to run on every PR - Fragile to timeouts in CI - Hard to debug when they fail (so many steps) Is the right answer to break them into smaller, more focused tests? Or accept that some critical paths need long tests and just manage them separately? What's the longest single test you'd tolerate in a healthy suite?