Which is probably why they dropped the prod db.
But I'm sure you can enlighten me with some actual research-based facts?
'Ship fast and break things' is one of the best and worst cultures in the tech industry.
You could argue that dev-ops model include automation for operations on production systems, with direct access to production systems limited to a reduced set of staff. Developers can create the change-sets to modify infrastructure, but those change sets are reviewed, validated, tested, and then executed on production via CI/CD Automation.
Infrastructure as Code, and Immutable Infrastructure lends itself well to that approach.
As part of a team (and an organization) that practices "devops" heavily, all of our devs do ops. We maintain the separation using an oncall/ops rotation and only touching prod from designated "ops" hosts. We also follow a "2 person" rule when touching production. We use a secrets management system to deliver credentials to hosts and alarming setup when the "wrong" hosts (e.g. dev hosts with access to prod creds) have access to certain creds.