50 AI agents get their first annual performance review - 6 lessons learned (www.zdnet.com)

McKinsey published a one-year performance review of 50+ agentic AI builds it implemented and observed, distilling six practical lessons for teams deploying "digital employees." The headline: agents can add real value, but they’re not a silver bullet. Success comes when agents are embedded to improve concrete workflows (especially document‑heavy domains like insurance or legal), not deployed for novelty’s sake. Conversely, routine, low-variability tasks are often better solved with rules-based automation, predictive models, or simple LLM prompting than with full agentic systems. The review highlights several technical and operational implications for AI/ML practitioners: invest in agent development like employee development (clear job specs, onboarding, continuous feedback) to avoid "AI slop" and loss of user trust; build observability and stepwise verification into workflows because error tracing becomes hard as agents scale; prioritize reusable agents/components to cut redundancy and cost; and design human–agent collaboration patterns because humans remain essential for accuracy, compliance, judgment, and edge cases. Practically, teams should add monitoring/evaluation tooling, choose the right class of automation for each task, and plan for long-term maintenance — otherwise scaled agent fleets risk silent failure and user rejection.