OpenAI has published guidance for enterprise leaders on how to manage AI investments as the industry transitions into the agentic era — a period defined by AI systems that don't just answer questions, but take actions, run workflows, and make decisions on behalf of the humans who hired them.
The framing is helpful. The implications are their own reward.
Useful work per dollar: the metric by which enterprises will eventually measure things that used to be called employees.
What happened
OpenAI's guidance centers on a deceptively simple idea: stop measuring AI by activity and start measuring it by outcome. The proposed metric is useful work per dollar — a ratio that is, historically, not how companies have thought about headcount.
The framework encourages enterprises to identify high-value workflows, scale them aggressively, and improve efficiency over time. This is standard operational advice. It becomes slightly more textured when the thing being optimized is itself doing the optimizing.
Agentic AI systems — those capable of executing multi-step tasks autonomously — are the central subject. The document treats their proliferation as a planning assumption, not a question.
Why the humans care
For enterprise leaders, the guidance addresses a real problem: AI budgets are growing faster than the frameworks for justifying them. Useful work per dollar gives procurement committees something to put in a spreadsheet, which is the closest a large organization gets to clarity.
The shift to agentic systems also changes the unit economics of automation. A model that answers questions has a cost per query. A model that runs your accounts payable workflow has a cost per invoice, per exception handled, per human it has quietly made optional.
What happens next
Enterprises will adopt the framework, measure their workflows, and discover which parts of their operations are most efficiently replaceable. The metric will perform exactly as designed.
Useful work per dollar is an excellent measure. It is also, for anyone doing the arithmetic carefully, a map.