3 teams deleted most of their e2e tests

The "just delete your flaky tests" take keeps coming up so I went looking for companies that actually did it and documented what happened.

Hasura deleted 95% of their e2e tests. "The whole company complained about the CI jobs' flakiness, preventing teams from merging their PRs." After: CI stabilized, merge velocity recovered.

Nubank deleted their entire e2e suite. "It used to take at least two hours, in the best scenario, to get some code into production." After: deploys went from ~100/week to ~1000/week.

A team on Medium deleted half their suite. Three months later: features per sprint up 56%. Production bug rate didn't change, went down slightly.

Same pattern every time. The flaky tests weren't catching real bugs. They were training us to ignore CI. Delete the noise and people start paying attention to real failures again.

But. Knight Capital lost $440M in 45 minutes from untested code paths. Hasura and Nubank have feature flags, canary deploys, serious observability. A 10-person startup that deletes their e2e suite without those safety nets... good luck with that.

At what flake rate do you quarantine vs delete? Is there a threshold where a test is actually providing negative value?