Evaluation, observability, control and recovery table stakes

ROI challenges highlight need for telemetry: 94% Day 2 issues, 72% higher ops costs. Anthropic MCP, NVIDIA AI-Q/Verified, Boomi Control Tower, Versa zero-trust, LOOP replay. ACC agent trajectories for long-context training; new eval frameworks. AAIF standards body (43 new members); auth platforms for MCP (WorkOS, Stytch, Auth0, Composio); SailPoint Claude compliance API; Torq/Jit $50M security context graphs. Anthropic Mythos vulnerability discovery. Google open-sources Agent Executor for durable execution and checkpointing.