There was a comment on HN which I keep thinking about. What is the value of expensive testing when you are promptly alerted of critical problems and can deploy in under a minute? Obviously not applicable for every scenario, but at this point I think for most non critical systems a good rollback/plan B is at least as important as testing.