Open Weights Hit the Frontier, AI Agents Get Real, Security Goes Mainstream — Weekly Synthesis, Week 31 2026
Seven days, three shifts: open-weight models reached frontier parity with closed labs on cost and capability, AI agents started doing real production work in the open, and AI security incidents stopped being a vendor-only problem and became a board-level one. Plus: physical AI shipped in volume, the White House carved out a regulatory lane for US open models, and the deeper adoption lesson crystallized around memory, verifiability, and the 'agency rupture.'