He was a PM, so he doesn't understand the real problems inside Google. Talking about agility: you have to provision TPUs years ahead. Developer jobs are taken over by higher priority production tasks all the time. Datacenters are always turned down, and job dependencies on datacenters are still managed manually, as the internal tools don't support automatic migration. Forgetting to provision necessery network quota means that even when you get a lot of GCUs, you have to wait months again to get network quota. If you find the right SRE though they try to be helpful, but their tools are limited as well.

I'm not saying things are bad though. There are lots of great tgings at Google, but agility is not one of them.