Past Adoption, Into Accountability: What Operators Cannot Yet Measure or Control

Operators have moved past whether to adopt AI and are now struggling to measure its value, control its cost, and govern the volume and access it creates.

Senior operators across consulting, software, franchising, technology, and regulated industries described a shared shift. The question is no longer whether to use AI but how to prove and control it. Several are building AI agents yet cannot measure impact, falling back on biased self-reported time savings to justify spend. Compute budgets are shrinking under real usage, forcing quiet rationing even where leadership mandates adoption. A consultant captured the double edge: AI is eroding a third of her firm's service revenue while she uses it to replace her own lawyers, feeding NDAs into chatbots without knowing if the output is sound. Operators also reported a swing from underuse to overuse, generating slop, while quality-control capacity has not kept pace with faster production, pushing cost and risk downstream. Technologists said AI now writes more code than humans can review, prompting homegrown controls like adversarial cross-model checking. The most consistent alarm was agentic governance: agents granted broad access to sensitive data, and staff clicking approval prompts mindlessly. The breakout group converged on the same themes, concluding that trial and error is no substitute for AI policy, and that deciding upfront how much authority to delegate to AI is now essential. For executives, the message is clear. Value measurement, cost control, review capacity, and access governance now matter more than adoption speed.