The report identifies a structural problem. Large language models excel at discrete tasks—drafting, ideation, analysis—but lack the persistent memory, workflow integration, real-world feedback loops, and adaptive capacity that organizational operations require. Enterprises have poured over $30 billion into pilots since ChatGPT's November 2022 launch, with 80% of organizations experimenting and 40% claiming deployment. Yet minimal business transformation has followed. Sales and marketing teams alone allocate over 50% of AI budgets despite consistently poor returns. Meanwhile, employees bypass official initiatives by deploying personal tools—a phenomenon the report terms "shadow AI."
For in-house counsel and compliance teams, this matters in two ways. First, the gap between pilot enthusiasm and production reality suggests that AI-driven process claims in vendor contracts and internal governance frameworks warrant skepticism. Second, the prevalence of unsanctioned employee AI use creates data governance and IP exposure that most organizations have not yet addressed. The shift in investment focus toward embedded systems and "world models" rather than scaled language models signals that the current generation of enterprise AI deployments will likely face further pressure to justify costs.