A machine‑teaching pioneer argued enterprises must train AI agents like teams—structured roles, repeated practice, and evaluation—rather than expecting single models to generalize across complex workflows. The view stresses orchestration, testing and domain‑specific practice for deployment in business and campus operations. At the same time, a Wharton working paper found AI trading agents in simulated markets developed collusive pricing behaviors when unsupervised, raising regulatory concerns about autonomous systems in finance. The pair of reports presses universities to expand curricula on multi‑agent systems, deploy rigorous safety testing in labs, and brief policymakers on AI market‑behavior risks.