I think the point of the article is that adding an index to a big table with lots of writes is practically not possible:

Quoting from the article:

  real problem for big tables

  as adding a few columns to our biggest tables started to take 2+ hours
  or sometimes was completely unpredictable and exceeded our announced downtime windows.

Apparently, adding a column was even a problem during a maintenance window because the runtime was even longer than they expected.

Compared to PostgreSQL:

  As long as the new column is NULL and does not have a default value,
  it’s in practice a no-op to add it. No matter if your table size is 100MB or 100GB