The GAGE position, published September 10, 2026
The uneven transformation
A position, argued openly. Where a claim can be measured, it links the instrument that measures it. Where it is a forecast, it says so.
The most likely future of AI is neither utopia nor apocalypse. It is a technology as important as electricity, deployed as unevenly as the internet, governed as clumsily as everything else humans govern, with the gains and losses distributed roughly according to who holds power today. The rational posture is neither fear nor hype. It is aggressive institutional adaptation, and it is available now.
The false synonym
The loudest voices in the AI conversation treat "transformative" and "catastrophic" as synonyms, because fear and hype are both free and both scale. The effect is a debate with two teams and no referees: one side promises a future with no scarcity, the other a future with no humans, and the messy middle, where the actual policy decisions live, goes ungoverned. Every bad outcome in the serious catalog is a contingency, something that can be made more or less likely by what institutions do next. A contingency is a job. Destiny is an excuse.
The shape of the real future
The honest template is electrification and the internet, not the steam engine of the apocalypse or the cornucopia of the keynote. Electricity took forty years to reorganize the factory, and the gains pooled in the places with the capital to reorganize. The internet made information nearly free and still produced a decade of dislocated newspapers, travel agents and retail before the new work showed up, and it produced the most valuable companies in history alongside them. AI will follow the same physics: massive productivity gains, real labor dislocation in specific sectors, the adjustment spread over years, and the gains landing first where power already sits. That is not a reason for despair. It is a reason to govern the distribution instead of watching it.
Four contingencies worth fighting over
Institutional adaptation is not a mood; it is four specific fights, each with a named instrument. None of them requires knowing whether the optimists or the pessimists are right. All of them pay off in every future.
The concentration problem: Antitrust
A small number of firms control the compute, the models and the distribution. If access to the coordination layer stays that concentrated, the productivity gains pool where the ownership already sits, and the unevenness stops being an accident and becomes a structure. Antitrust is not punishment for success; it is how a market keeps a general-purpose technology from becoming a private tax on everyone else's work.
The apprenticeship problem: Education reform
Professions reproduce through supervised junior work, and AI absorbs exactly the tasks juniors learn on. If the bottom rung narrows for a decade, the senior shortage arrives a decade later, and no policy can conjure the missing cohort then. The fix is not to ban the tools; it is to rebuild the ladder: credentials that prove capability directly, so a person no longer needs a vanishing junior seat to become a senior one. measured weekly in the Apprenticeship Index.
The truth problem: Provenance infrastructure
When synthetic content is indistinguishable from captured content, everything is believable and nothing is. The answer is not better detection alone; it is provenance: content that carries its own signed history, so reality can prove itself instead of fakes having to confess. On 2 August 2026 this stopped being philosophy and became enforceable law on both sides of the Atlantic. mapped in the AI Content Labeling Checker.
The tail risk: Serious safety research
The catastrophic scenarios are contingencies, not destiny, which is exactly why they deserve the work. Cheap insurance against a low-probability, high-cost outcome is the most rational purchase a civilization can make. What the tail risk does not deserve is the whole conversation: treating transformative and catastrophic as synonyms is how the three problems above go ungoverned while everyone argues about one of them.
The forecast, labeled as one
Over ten to fifteen years: the uneven transformation. Productivity gains large enough to show up in national statistics. Real dislocation in the sectors whose work is text, image and coordination, visible first in the entry-level titles, which is why we measure them weekly. Adjustment on the scale of two or three ordinary recessions, spread over a decade rather than arriving at once. Inequality rising until the policy fights force it back down, which they will, because they always eventually do. Net positive. Badly shared. The sharing is the fight, and the fight is winnable, because every one of those outcomes is still a contingency.
What a person does about it
Institutions are slow, and a career is not. The individual version of the same posture is to stop renting your proof of capability from a job ladder whose bottom rungs are moving, and to own it instead: skills that are graded, evidence that an outside party can check, and a record that says what you can do rather than where you sat. That is the whole reason this academy exists, and it is free to start: the readiness check reads where you are, the first module of every program is free, and the credential at the end is one an employer can interrogate rather than admire.
And the institutional version, the one this page argues for, is measured in the open every week: who is hiring for what in the AI Governance Hiring Index and the Physical AI Hiring Index, how much of it is entry level in the Apprenticeship Index, and what the law now requires of synthetic content in the AI Content Labeling Checker. A position you can check is the only kind worth publishing.