The Welder, the AI Module, and a Plan Still Being Drafted

The welder's hood snaps down early in the morning, well before orientation starts and before the printed curriculum packets have been counted again by the coordinator. A knot of apprentices stands around the booth, boots on the yellow line, nitrile gloves still bunched in pockets, breath visible under the roll-up door. The one who got there earliest is already asking what the AI module in week five is going to cover, because his cousin at a plant two counties over has started one and he wants to know whether this one will be the same. The coordinator says the curriculum is in draft. He does not say that the draft is national.

That answer, in community college CTE labs from Mobile to Fort Payne, is a report on where the federal workforce apparatus is sitting this October. Alabama's Commission on Higher Education received a $12.5 million competitive federal grant last autumn to catalogue existing apprenticeship frameworks, build a machine-readable training-outline tool, and reduce start-up costs for new sponsors. Colorado's Department of Labor and Employment received the same amount on the same day, scoped to pre-apprenticeship conversion and high-school pathways. Secretary of Labor Lori Chavez-DeRemer tied both awards to a target the administration has been quoting since spring: one million new apprenticeships. Thirteen months later the department has not published the plan it was asked to write to get there.

Note on the April 1 solicitation

On 1 April 2026 the Employment and Training Administration posted Solicitation 1605C2-26-R-00003, "Advancing Artificial Intelligence Skills in Registered Apprenticeships," with a one-year base period and four option years, to be awarded to a single national intermediary. The three priorities listed in the release are embedding AI training in existing programs, bringing apprentices into roles that build or apply AI systems, and strengthening pipelines in data centres, telecommunications and advanced manufacturing. Deputy Secretary of Labor Keith Sonderling called the Registered Apprenticeship a highly effective workforce model and a central strategy for meeting projected demand. The release did not give a contract value, an enrolment target, or a proposal deadline. Four weeks later the department added more surface area to the same architecture by launching an AI in Registered Apprenticeship Innovation Portal during National Apprenticeship Week, as Weslan Hansen reported for MeriTalk on 2 April 2026, with training modules by industry and three paths to add AI content to existing or new programs, and sitting alongside a $145 million January 2026 expansion commitment and a March 2026 Make America AI-Ready text-literacy pilot. None of those press moments sat next to the plan the executive orders had required.

What CSET found missing

That plan is the piece of paper the Center for Security and Emerging Technology was looking for when, on 20 November 2025, analyst Ali Crawford published a progress note on Executive Orders 14277 and 14278 and found the 120-day apprenticeship-strategy deadline had closed in late August 2025 with no public plan filed. The Departments of Labor, Commerce and Education did publish America's Talent Strategy on 12 August 2025, a five-pillar document, as Crawford describes it, naming demand-driven strategies, worker mobility, integrated systems, accountability, and flexibility-and-innovation. The strategy's only numeric claim is "hundreds of thousands" of jobs. The one-million line still lives in press releases, not in a filed plan. Section 8 of Executive Order 14277 also required the Secretary of Labor, within 120 days of signing, to set specific goals for growing AI-related Registered Apprenticeships, issue WIOA guidance encouraging youth funds for AI skills, and clarify governor set-aside use. The goals paragraph is one of the ones Crawford's progress note did not find written down in public.

The numbers underneath

The structural numbers under the one-million figure are not inferred. Kristina Rosario's May 2026 brief for Jobs for the Future counts 702,000 active Registered Apprentices in fiscal year 2025, up from 396,000 in fiscal 2016, with year-over-year growth that has fallen from above ten percent before the pandemic to about five percent now. On the current trajectory the system reaches roughly 850,000 by fiscal 2030, not a million. The brief estimates 150,000 to 160,000 cancellations against about 120,000 completions each year, and reports that around 82 percent of pre-apprenticeship participants never enrol in a registered programme at all. Rosario proposes a 5-year investment strategy, a public-private partnership with industry seat commitments, state growth and innovation zones, expanded pre-apprenticeship and degree-apprenticeship models, and federal-contractor requirements as the architecture needed to close the gap. The contracting notice from April touches none of those pieces.

The money the department has moved, and what it does not fund

What the department has moved, in parallel, is a sequence of smaller grant lines. Nearly 84 million dollars went out in formula and competitive State Apprenticeship Expansion grants in June 2025. The 25 million in combined Alabama and Colorado awards followed in September. The Industry-Driven Skills Training Fund opened on 11 August 2025 with 30 million dollars in outcome-reimbursement awards up to 8 million per state agency. And Training and Employment Guidance Letter 15-25, issued 25 June 2026, made 50 million dollars available in Rapid Reskill Employment Recovery National Dislocated Worker Grants, individual awards from two to eight million. None of these lines is small. Together they add roughly 189 million dollars of flow against a target that would need, by the Jobs for the Future arithmetic, to add about 150,000 net active apprentices on top of the current trajectory in five years. The flow is going where the department can obligate it quickly, not where the attrition arithmetic eats the enrolment number.

Reading the Deputy Secretary's defence

Sonderling's defence of the architecture is worth taking seriously on its own terms. The Registered Apprenticeship is a genuine piece of workforce infrastructure, built around a tripartite relationship between the apprentice, the sponsoring employer, and the Office of Apprenticeship, with wages that rise against competencies rather than against seat time. A system producing 120,000 completions a year inside a labour market this large is doing real work. The claim the Deputy Secretary is defending, that AI-skills content can be added on top of a model that already runs, is correct in its narrow form. The claim it does not cover is whether that model, left to the growth curve it is on, can carry a million apprentices by 2030. On that question the April 1 contracting notice is silent, and the national plan the executive order required would have been the right place to answer it.

The ratio and the sponsor

The gap sits in the places the money is not going. Employer-sponsor acquisition is a sales process. It costs a state workforce board, by the operating estimates most state apprenticeship offices run from, somewhere between fifteen and thirty thousand dollars of staff time per net new sponsor, and the national intermediary the April contract is procuring is not funded to pay that cost at scale. The sector pattern makes the problem concrete. Construction is still the largest employer of Registered Apprentices in the United States, by Rosario's count. Manufacturing has fewer than 30,000 apprentices against several million workers, which is roughly the ratio a growth target of this size would have to invert. Advanced manufacturing is one of the three pipelines the April 1 solicitation names, and the pipeline it names is the one with the thinnest existing sponsor base to build onto.

Journeyman-to-apprentice ratios are set in a mix of state licensure code, employer collective bargaining and industry practice. The AI-skills contract touches none of those files, because none of them is a curriculum file. Retention, the leak at week six and the longer leak at month eighteen, needs a mentorship architecture that costs money and lives inside the sponsor relationship, not inside the training provider's module library. None of this is a mystery. All of it is what a filed scale plan would have had to allocate.

The coordinator's answer, read against the law

There is a longer arc under this file. The last American administration that treated apprenticeship as a serious scaling object, rather than as a line in a budget or an executive-order citation, was the Johnson administration, and even then the Manpower Development and Training Act produced numbers under half what the department now claims it will exceed. Between then and now, every half-decade produced a target, a strategy document, a press event, and a growth curve that bent less than the target required. The welder's hood will come down tomorrow morning the same way it came down this morning. The curriculum packet will be a little thicker, with a new week-five module in it. The coordinator will still answer that the plan is in draft. Somewhere above him, in a building where contracting notices are drafted and five-pillar strategies are launched, the plan the law required a year and a half ago is still, in the plainest reading of the word, not written.

Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan for Energy AI startup, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.