Introduction: The Missing Returns

Global corporate investment in AI reached $581.7 billion in 2025, according to Stanford HAI's AI Index, a 130% increase over the prior year. Over the same period, PwC's 29th Global CEO Survey asked 4,454 chief executives what that class of spending had done for their businesses. Fifty-six percent reported zero financial impact from AI. Twelve percent reported benefits to both cost and revenue.

The models are strong and visibly improving; anyone who has used a frontier LLM (large language model) knows the capability is real. What those numbers describe is capability failing to convert into returns, at extraordinary scale, across nearly every industry at once.

When a gap this wide persists this long, the interesting explanations are structural. The whitepaper this guide draws on makes a specific structural claim: the missing piece is not a better model but a missing architecture. There is currently no standard for what an owned AI deployment looks like, no standard unit in which AI work can be packaged, exchanged, and reused, and no container in which an organization's accumulated context, the thing that makes AI output specific and valuable rather than generic, can be captured, governed, and shared. Absent those standards, organizations rent intelligence from vendors, and renting has costs that compound quietly until they show up in a CEO survey as "zero financial impact."

That claim has three parts, and this guide examines each: a diagnosis (Part I: what renting actually costs, and what asset is leaking), a design (Parts II and III: the architecture the paper proposes, from context artifacts through validation to sovereign deployments), and a forecast (Part IV: the economy the paper predicts will form once AI work becomes ownable and exchangeable, and the open-standards argument for why that economy compounds).

A note on posture. The diagnosis is well evidenced and largely third-party. The design is coherent, and pieces of it exist in the open-source ecosystem today, but it has not been proven at market scale. The forecast is a bet, and this guide will name the conditions under which the bet pays off or fails, because a field guide that cannot tell you what failure would look like is a brochure.