15-20 years ago, you'd run a tightly-coupled monolith in a colo and do scheduled downtime every few weeks to deploy everything. Everything was more error-prone

15-20 years ago, you'd run a tightly-coupled monolith in a colo and do scheduled downtime every few weeks to deploy everything. Everything was more error-prone and fixing errors could mean a physical trip to the colo or downtime in the middle of customers' workdays. We probably have the same amount of seat-of-pants, spit-and-bailing-wire problem-solving happening today, but the problems have moved.