A scorecard for the AI age

OpenAI says businesses should judge AI by work completed, not just by adoption metrics such as seats or active users. In a post published on July 17, 2026, the company argues that the best measure is whether AI helps produce useful outcomes faster and at a lower total cost. The article introduces a framework it calls “Useful Intelligence per Dollar.” It looks at four areas: whether AI is completing meaningful work, what a successful task costs, how dependable the output is, and whether the value improves as usage scales. OpenAI says this approach is more useful than measuring only token prices, because real business costs also include retries, human review, and time spent correcting errors. The post gives examples from support, engineering, legal, and finance workflows, and says newer model tiers are designed to balance speed, cost, and reasoning depth. It also emphasizes that reliability, security, privacy, and human oversight are important as AI systems move from drafting to taking action in business processes.