My first game day at the new badge ran Thursday. Yes, week two. The fastest way to learn a platform is to watch it degrade on purpose, and the team — bless them — agreed. We deliberately blackholed our primary model provider in staging, and then, with sign-off and new-guy audacity, for eleven minutes in production.
The fallback worked. Routing shifted traffic to the secondary and the self-hosted tier inside 40 seconds, and the golden test sets confirmed quality stayed within tolerance. The surprise, as always, was adjacent: our status-page automation composes its incident notices with the same model provider we had just blackholed. It tried to announce the drill using a model that wasn’t there. The room’s laughter was the sound of tuition being paid at drill prices instead of outage prices. The fix — plain static templates for incident comms, paper runbooks for the AI era — shipped Friday. The drill is the deliverable. It has been the deliverable for thirteen years and one new layer of stack.
The pager has a colleague now. Every incident-management vendor’s 2026 trend report converges on the same architecture: AI as first responder — alert triage, dedup, context assembly — with humans promoted to reviewers of the response. Both hands, per doctrine. The triage layer is genuinely good: our own deployment cut page noise about 40% last quarter, and the disk-full page that ruined my sleep in 2013 would today be diagnosed, contextualized, and drafted-for before a human woke. But the new failure mode is confidence: a responder that mis-triages confidently teaches the on-call to trust it at exactly the wrong moment. Automation bias is the alert fatigue of the 2020s. So our rotation now includes a weekly ritual: audit the robot. Seeded fake anomalies, measured catch rates. The watchmen get watched; trust gets earned by being graded.
Elsewhere: the EU’s financial-sector incident reporting passed its first anniversary and became routine. The Premier League’s winter fixtures delivered their annual variance sermon, and Null Pointer Exception’s postmortem of its own fantasy-football simulation concluded, in finest blameless tradition, that the model was fine and the Fantasy PL was flawed. The Winter Olympics open in fifteen days — next entry files from the eve.
TIL: aviation solved this decades ago. Crew-resource-management research on “monitoring the automation” — say out loud what the system is doing, run mandatory disagreement drills, calibrate trust per-function rather than per-system. Ops keeps inheriting aviation’s homework a generation late. Still better than never.