I find it interesting that this reflection didn't mention SLI/SLO/error budgets, which Google SRE has championed for a long time.
My impression is that they're nice in theory, but less useful in practice. I'm yet to see an error budget effectively inform eng decision making.