OpenAI's 'Five Principles' for Managing AI Investment in the Agentic Era

OpenAI has published an article explaining how businesses should manage AI investment as workflows become longer and more complex in the “agentic era.” It argues that value can’t be measured by per-token pricing alone, and companies should instead focus on “useful outcomes per dollar.”

Details

The article lays out five principles for business leaders and IT administrators.

• Strengthen visibility: Understand who is using which model or product, and for what purpose. A redesigned Admin Console enables breakdowns by user, team, product, and model, helping distinguish waste, pilot programs, and mission-critical work
• Evaluate ROI based on outcomes: Rather than picking the cheapest model, evaluate the total cost including failures, retries, and reviews. Track “cost per approved outcome” (such as support tickets resolved or code changes tested) alongside business metrics like time saved or risk reduced
• Make governance the foundation: Establish governance before scaling. Define what data ChatGPT can access, which tools it can use, approval requirements, and resource allocation (especially important when using plugins, connectors, or Computer Use features)
• Allocate investment like a portfolio: Invest across three tiers — “everyday productivity gains,” “repeatable workflows for specific tasks,” and “strategic investments that leverage proprietary data” — to support a progression from exploration to validation to production
• Match capacity to needs: Rather than having each team build its own infrastructure, scale proven workflows through the appropriate product, such as ChatGPT Work, OpenAI Frontier, or a Deployment Company

How to try it

The article highlights the redesigned Admin Console (with spend management features), OpenAI Deployment Engineers, and a zero data retention option. Enterprise administrators can start gaining visibility into usage right away through their existing Admin Console.