The New AI Economics Standard Measures Labor Replaced

← Back · ACT III · Chapter 22: Embedding Buyer Governance

On 4 August 2026 the Linux Foundation launched the Tokenomics Foundation, a vendor-neutral body chartered to establish open industry standards, benchmarks and best practices for the economics of AI. Thirty initial members, per the Linux Foundation, including Accenture, BNY, Broadcom, Hitachi, IBM, JPMorganChase, Lenovo, Oracle, SAP, ServiceNow and SHI. The governing board convened on 30 July.

The largest open-standards institution in technology looked at unproven AI spend and did not propose an office, a role or an engagement. It proposed published definitions, a shared reference model, common measurement and a certification, governed by a board rather than by any single company. That corroborates an argument already made here in Why Offices Set Up to Track Purchase Results Keep Getting Cut, and corroboration from that institution is worth more than another round of assertion. Take it plainly. Then read what the standard counts.

What The Foundation Concedes

Jim Zemlin, Executive Director of the Linux Foundation, in the launch release: “Tokens have become the commercial expression of the entire AI economy, yet there is no shared way to connect AI spending to value.”

That is a concession from the center of the industry. After three years and several hundred billion dollars of enterprise commitment, there is no shared way to connect spending to value. That is a different complaint from a shortage of dashboards or immature tooling, and a harder one to fix. J.R. Storment, Executive Director of the new Foundation, states the consequence: “Every CEO is being asked to show returns on all of that without a shared way to count it.” Every diagnosis in that release is a diagnosis of a missing standard, and the remedy proposed matches the diagnosis, which is rarer than it sounds.

The Opening Definition Of Value

The roadmap set out in the launch release names value measurement as one of its initial pieces, and the wording deserves attention: “A framework for relating AI spend to outcomes, starting with the share of work completed without human involvement, measured against what the process costs today.”

Read it as an equation. The numerator is work finished without a person. The denominator is what the process cost before. It is a ratio of labor displaced against prior cost, and it is a measure of substitution. It answers how cheaply the work got done. It does not answer whether the work was worth doing, whether the result the spend was authorized to produce arrived, or whether the buyer is better off than before the money moved. An organization can raise that ratio every quarter while the outcome named in the business case never appears, because cost per unit of work falls and nobody is required to revisit the promise that released the budget.

There is a popular answer to this and it is the wrong one. Through July and August the argument that tokens are the wrong yardstick has become common, and its remedy is uniformly to pick a better metric: workflow return, cost per business result, outcome attainment. That keeps the problem where it was. A better metric chosen by the same party, drawn from the same telemetry, still measures the supplier’s side of the transaction. The question is not which number to use. It is who has standing to write the definition the number reports against.

A Standard Counts What Its Members Can See

The obvious objection is that suppliers are writing the buyer’s measure. That objection is wrong and worth retiring before it spreads. Membership includes large buyers of AI as well as sellers of it, governance sits with a neutral foundation, and one founding member says the point outright. Bill Lobig of IBM: “No single provider should define those benchmarks.”

The sharper objection is that seating real enterprise buyers therefore solves the buyer-authority problem. It does not, and the reason is structural rather than motivational. A standard is built where the data is. Tokens, calls, compute, cache and cost per unit of work are all recorded by the systems these members operate, so those are the quantities a shared specification can define, benchmark and certify. Seats do not move data. Realized outcome is recorded somewhere else entirely: in the buyer’s own business, in the retention that held, the margin that moved, the cycle that shortened, the claim in the business case that authorized the purchase. No supplier’s telemetry contains any of it, and no specification written from supplier telemetry will reach it, whoever is in the room.

Laurent Gil of Cast AI named the shape of this in the same release: “The industry is standardizing the meter in the middle and ignoring both ends.” His next sentence matters more: “Cast AI joined the Tokenomics Foundation to help measure both.” A founding member has conceded the diagnosis and drawn the opposite conclusion, that the far end is merely unmeasured and can be brought in. Unmeasured is a gap that effort closes. Unreachable is a property of where the record lives, and effort does not close it.

What The Buyer Still Has To Write

Cost standards can travel from the supply side, because the supply side holds the meter. The narrower version of that point, about a single instrument, is Vendor Total Cost of Ownership Models Miss Your Biggest Costs; the Foundation raises it to the level of an industry specification, where it binds harder. Neutrality is not the missing piece either. A specification can be openly governed and still be written from one side of the table, which is the distinction drawn in An ‘Open’ Standard for Value and Pricing Models Is Vendor-Run.

What the buyer has to write is not a measurement problem. The claim is stated before the money moves. The measure is fixed at authorization rather than assembled at the review. The terms are published, so the same definition is in front of everyone who will later argue about the result. And the record outlives the people who made the decision, because value is settled long after the sponsor has changed roles, which is the argument of Every Executive Claims to Sign Off on AI Purchases. None of it depends on a new seat, a new office or an advisory engagement. It depends on the buyer treating its own definition of value as a published term of the purchase rather than as a slide.

The Tokenomics Foundation will produce something useful, and the field is better for it existing. It will tell an enterprise, in a common language, what its AI cost. Its own draft definition, version 0.5.2 of 1 September 2026, now ends on realizing measurable business value, which is the right ambition and does not change where the record sits. What the money was for is still written in the buyer’s own business, and no member’s telemetry reaches it. The other half of the ratio is still unwritten, and the buyer is the only party with standing to write it.

A standard for the spend answers what AI cost. Only the buyer can set the standard that says what it was for.

Governance Check: when your AI spend is next reviewed, whose definition of value will the review use, and where was it written down?

Sources: Linux Foundation press release, “Linux Foundation Launches the Tokenomics Foundation to Define the Economics and ROI of AI Value,” 4 August 2026; all quotations and the value-measurement wording are verbatim from that release. The release states thirty initial members and names twenty-nine. Tokenomics draft definition v0.5.2, tokeneconomics.com, 1 September 2026. Launch coverage: The Stack, 5 August 2026, and Tech Times, 4 August 2026.

Keep reading: ACT III · Chapter 22: Embedding Buyer Governance