“Boris Cherny predicts that within a year, AI models will be significantly better aligned, reducing the need for product-level safety mechanisms like prompt injection defenses, static command verification, and mandatory human-in-the-loop approvals.”