On July 13, the Stanford Digital Economy Lab released We Must Act Now: A Statement on AI's Transformation of the Economy. Sixteen Nobel Laureates and more than two hundred economists and AI researchers signed it.
The statement argues that AI may drive an economic transformation larger than the Industrial Revolution, on a vastly shorter timeline, and calls on economists, policymakers, and technology leaders to build the policies and institutions needed to steer it.
It was organized by economists at Stanford, the University of Toronto, the University of Virginia, and METR. Signatories include multiple Nobel laureates in economics, several of the field's most cited labor and growth economists, Turing Award winners in AI, and researchers from every frontier lab, including OpenAI, Anthropic, Google, and Microsoft Research.
The declaration is right about the stakes and correct about the direction. It also stops at the door of every company actually shipping AI right now.
The gap between statement and system
Guardrails do not exist in policy papers. They exist in software that runs when an agent makes a decision. Institutions do not steer AI. They set rules that get enforced, or ignored, by the systems companies operate day to day.
If a customer support agent enters a retry loop overnight and burns $47,000 dollars, no economist writes that check. Finance does. And Finance has almost no visibility into which workflow, which customer, or which feature caused the spike.
The macro conversation about AI economics has a matching micro problem. Companies cannot govern what they cannot attribute. Most of them cannot attribute anything below "the AWS bill went up."
Where the economic control layer actually lives
Revenium exists because the guardrails the statement calls for have to be operational, not aspirational. In practice, that means three things:
Attribution at the transaction level. Every model call, agent action, and tool invocation tied back to the customer, workflow, feature, and team that triggered it. Not sampled, and not aggregated at the account level. A complete economic ledger.
Cost-to-outcome visibility. Spending $40,000 on an AI feature is not a problem if it generates $200,000 in revenue. It is a business-ending problem if it generates $8,000. Without connecting cost to outcome, every AI investment is a coin flip dressed up in a dashboard.
Real-time enforcement. Alerts after the fact are not controls. A control system watches economic behavior as it happens, applies limits, and stops unprofitable usage before it hits the P&L. That is the difference between an incident alert and an actual circuit breaker.
What the statement gets right that the market is slow to admit
The signatories are calling attention to something the AI industry has mostly avoided saying out loud. AI is non-deterministic. A single query can trigger ten API calls or ten thousand. Traditional forecasting, pricing, and cost control all break.
You can’t govern a system whose economic behavior you can’t predict unless you build the governance layer into the runtime. Policy will not catch up in time, and regulation will not either. Neither will board-level reporting cycles that measure spend a quarter after it happened.
The 2026 handoff
2025 was the year companies shipped AI features because they had to. 2026 is the year they have to prove those features are worth shipping. Boards are asking for unit economics. CFOs are questioning AI budgets they signed off on twelve months ago with no framework to evaluate them.
We Must Act Now points at the macro, but the work still has to happen at the micro. Every company running AI in production is already inside the transformation the economists describe. The open question is whether it has the economic ledger, the attribution, and the controls to survive it.
Revenium was built for that layer. It runs at the transaction level, every second, inside your product, while the policy conversation plays out elsewhere.
