> However, if what you have to show for it is making the life of a hundred random startups easier, well why exactly should a Google promo committee care?

He is actually talking about this in his tweets: "fixing a bug in kube-apiserver might retain a GCP customer", now he can put revenue associated with that customer into his perf packet?

The problem is that's a very strenuous argument to make (unless he could find an email that said that a customer was going to leave unless they fixed the bug) vs another engineer who found a new way of compressing emails and save the company X exabytes of storage space.

Keeping the ship running seems to be continuously undervalued at these companies compared to how they value building a new wing.

FAANG manager here, with experience in other industries. I have definitely seen that promo committees here value maintenance work less than other industries. I think the problem is that the committee is made up of individuals from other functions, so they really have very little idea about the services a candidate supports, they're just looking at the accomplishments written in the packet. In other industries, that's typically but the case - if you run a service that has a reputation for being unreliable, you're not even getting considered.

You still have to do the above and beyond work elsewhere, but in FAANG some people will get into the conversation by ditching maintenance and focusing solely on the stretch work, and they tend to have success getting promoted.

> he could find an email that said that a customer was going to leave unless they fixed the bug

Yes, otherwise it can totally be that his impact is not that large actually.

Could be could not be, hence the hard to measure part.

I've stopped using many tools because they created just too many issues. And in that case you will not write these emails as a customer, you just leave (as long as there are better options).

There are a whole lot of companies who can look at MTTR in terms of dollars, especially eCommerce. Rollback failures and CPU throttling can translate very directly to dollars when each request is a potential transaction.