My Bullshit detector says that...
“In the old days, ... optimised for reducing Mean Time Between Failure [MBTF]. This was appropriate in a world with a high cost to repair, ... In the new DevOps world, you optimise for Mean Time To Remediate (MTTR), because you make the cost of repair very low by allowing very rapid deployment and limiting exposure,”
...is just our old favourite: "Beta test in Production". No surprise it comes from an industrial scale patch factory like MSFT. Poster children for this approach are IE, Java and Flash. Don't worry too much about shipping security holes because you've got high performance patch servers that allow you to fight a continual rearguard action against your own continuous delivery of agile defects.
The problem is 'cost of repair' isn't a simple metric. The cost to Adobe of shipping continuously insecure Flash is destruction of customer confidence in the brand, which is not something a DevOps toolchain can track and churn out pie charts about.
Cost of remediation is already way too low and has been encouraging bad practice for years. Lowering it further is solving the wrong problem and it only takes one serious economy updating it's Trading Standards and Consumer Protection legislation to invalidate the entire premise.
[*] And lets not even think about IoT with millions of units with embedded firmware that is next to impossible to update efficiently if at all.