Two news sites moved off a managed host and kept running by an AI engineer

Editors reported blank images and articles that did not appear on the home page. Uptime checks were green while readers saw errors. Backups existed somewhere, but nobody had ever restored one. Every change went through a ticket to an outside developer, and the managed host cost more each year.

The sites were migrated to their own AWS account with a CDN in front, and the Engineer took over their operation. It watches uptime from outside every five minutes, publish lag (is the newest article really on the listing pages?), PageSpeed weekly, and a security sweep weekly. The first sweep found a full database dump left publicly downloadable by the old host and removed it the same hour.

The Engineer has tiered rights: after a backup it may clear caches, restart services, roll back a plugin or block an attacker on its own. Code, database, theme and DNS changes are proposals: Clara asks in Slack, a person approves, the Engineer applies and reports the outcome. Every alarm, change and decision is on the Agent activities page.

Backups are hourly to object storage with point-in-time recovery, daily disk snapshots, a weekly copy off the cloud, and a monthly restore drill on a throwaway server, so "we have backups" is proven, not assumed.