Overview
- Recent reporting from Oct. 21–23 confirmed that 47% of workers surveyed already use agentic AI that can plan and execute multi‑step tasks, consult documents and coordinate actions.
- Industry surveys show heavy deployment plans but weak oversight: SAP LeanIX found about 98% of organizations have implemented or plan agents while only 17% have visibility into agent performance and compliance.
- Agentic flows cost far more to run than simple chatbots, with analysts estimating 5–30 times higher model inference usage and 73% of organizations reporting AI costs above projections, which threatens promised savings.
- Real harms are emerging because many systems act without human review, including a reported Australian case in which an automated evaluation triggered a written warning for a cashier with 99.8% accuracy.
- Experts recommend a people‑first sequence—map and redesign processes, establish single sources of truth, add agent identities and kill switches, require human checks, and invest in explainable or neurosymbolic methods to reduce errors and abuse.