The Capacity to Absorb the Damage

Jasmine Sun's NYT op-ed on the AI permanent underclass is the most honest documentation we have of what the frontier labs believe is coming. The labs know. The economists know. The political strategists running 2028 messaging know. The piece catalogs the awareness across every tier of the AI industry and finds a uniform answer: the median person is screwed and nobody knows what to do about it.

The piece is missing the part that determines whether any of the proposed solutions can actually run.

Sun's piece reports on policy ideas — OpenAI's Industrial Policy for the Intelligence Age white paper calling for a 32-hour workweek and a public wealth fund, Anthropic's gestures toward expanding the relational labor sector, David Shor's polling showing that a federal jobs guarantee polls well and UBI does not, Carl Benedikt Frey's reminder that "the short run can be a lifetime." Every one of these proposals, if it became law tomorrow, would land on a federal government that has spent the last twelve months deliberately reducing its capacity to execute large transition programs.

This is the part nobody is writing about. The transition has no operator.

The capacity gap is the policy gap

The federal workforce shrank by roughly 270,000 people in 2025. The acquisition workforce — the people who write, scope, and oversee the contracts through which any federal labor transition program would actually flow — is operating at vacancy rates approaching 40 percent at some agencies. 18F was eliminated in March 2025. The 10x program at GSA was effectively shut down. The Technology Modernization Fund pipeline is starved. The Centers of Excellence are hollowed. The Presidential Innovation Fellows program is gutted. The mid-career technical talent that any transition program would need to design intake systems, eligibility logic, payment infrastructure, and case management has been dispersing to the private sector at the rate of several thousand people per month.

A federal jobs guarantee at the scale Shor's polling implies would require standing up program offices in every state, hiring case managers, building benefits delivery infrastructure, integrating with state unemployment systems, and establishing oversight mechanisms. The federal government did this kind of work during the 2009 stimulus and during the COVID response. In both cases the execution was ragged, and in both cases the government was operating at full staffing with intact procurement infrastructure.

The current government cannot execute a transition program at that scale. The capacity does not exist. Cutting the capacity was the explicit project of the last twelve months and the cuts succeeded in their explicit aim. Now the same political coalition that ran the cuts is being told by the AI industry that a transition program of unprecedented scale will be needed within five years.

The labs' value-partition problem

The frontier labs are running a textbook value-partition operation, in the civil economics sense. Anthropic books $30 billion in annualized revenue selling the agents that displace knowledge work. OpenAI's growth trajectory is similar. The capture is private, immediate, and enormous. The displacement is external, dispersed, and lagged. The residual obligation to absorb the displacement lands on a public sector that has been deliberately weakened and on workers whose career ladders are being cut underneath them.

The white papers and the public statements are forms of pre-emptive boundary-drawing. The labs are establishing on the record that they identified the problem, named the policy responses, and called for action. Whatever the public ends up holding, they will be able to point to the warning. The transaction boundary closes at the white paper. Everything outside it becomes residual public obligation.

This is not a moral judgment. It is the value-partition pattern that any sufficiently large extractive industry produces under conditions of weak public counterweight. The fossil fuel industry did the same thing with climate. The opioid industry did the same thing with addiction. The frontier labs are running the playbook earlier and more articulately than their predecessors, but the structure is identical. Vp captured fast. Ve and Vr deferred onto institutions that did not negotiate the boundary.

The trajectory mismatch

The federal capacity recovery curve and the AI displacement curve are running in opposite directions on the same timeline. Every month the dispersed federal cohort gets further from reentry. Clearance currency lapses. Non-competes get signed. Institutional memory walks out of the building. The cost of rebuilding equivalent capacity from new hires runs five to ten times higher than rapid rehire of the existing cohort, and produces lower-quality results because the relationship layer cannot be hired for.

Every month the displacement steepens. OpenAI's GDPVal benchmark showed an 80 percent win rate against human professionals across 44 occupations within months of release. Block laid off nearly half its workforce in March and the stock surged 25 percent. Anthropic's own research finds that junior engineers using AI agents complete tasks no faster than peers and understand their work less afterward — meaning the entry-level cohort is being deskilled at the same time the entry-level positions are being eliminated.

The window is closing in measurable increments at both ends simultaneously. The capacity to run the transition is shrinking. The need for the transition is growing. The gap between them is the entire policy problem and almost nobody is naming it as such.

What the form layer cannot fix

The Public Mechanics frame is useful here. Public services run on a chain: policy → form → encounter → consequence → feedback. The form layer is what most policy discourse treats as the work — the white paper, the bill text, the announcement, the press conference. The encounter layer is where the citizen actually meets the program. The consequence layer is where outcomes register. The feedback layer is what tells the system whether it is working.

The labs, the think tanks, and the political strategists Sun documents are operating almost entirely at the form layer. Industrial policy white papers. Federal jobs guarantee polling. 32-hour workweek proposals. Public wealth fund concepts. None of this work is wrong. All of it is necessary. None of it is sufficient.

The encounter layer for an AI labor transition program does not currently exist. There is no agency tasked with delivering it. There is no procurement vehicle scoped for it. There is no case management infrastructure, no eligibility system, no payment apparatus, no oversight function. The federal employees who would have built those systems are dispersed. The contracting officers who would have written the implementing contracts are absent. The mid-career technical talent that would have done the integration work is at Anthropic and OpenAI, ironically enough, helping to accelerate the displacement that the absent transition program is supposed to absorb.

You can pass any law you want at the form layer. If the encounter layer cannot be built, the law does not produce outcomes. It produces press releases about outcomes. The Public Mechanics diagnostic is that form layer activity without an intact delivery chain produces performative concern without operational effect. That is what the Sun piece is documenting without quite naming it. The labs publish the form. Congress is captured by the money. The agencies that would execute have been hollowed out. The chain is broken at multiple points and nobody is fixing it.

The Shor problem

David Shor's pitch is operationally important and politically incomplete. He has correctly identified that AI labor displacement is the political opening of the 2028 cycle, that populist framing outperforms technocratic framing, that a jobs guarantee polls well and UBI does not. The polling is real and the messaging frame is correct. The pitch is for $700 billion in lab spending vs. less than one hour of that spending on Democratic political messaging — a true and useful asymmetry.

The pitch does not address what the politicians elected on this messaging would actually be able to do. Winning the 2028 election on AI labor displacement messaging produces a Congress and a White House with a mandate for a transition program and no functioning apparatus to execute it. The 2026 and 2027 cohort dispersal will be largely complete by January 2029. The acquisition workforce vacancies will have compounded. The institutional knowledge gap will be wider. The contractor capture of the rebuild will be further entrenched.

A campaign that sells transition policy without simultaneously selling capacity recovery is selling something that cannot be delivered. The political will to execute lands on a delivery system that has been disassembled. The voters who supported the messaging see the transition program fail at the encounter layer and reasonably conclude that the politics was hollow. This is how populist openings get squandered.

The capacity recovery work needs to start now, not in 2029. It needs a 10x-style micrograms program to rebuild internal delivery capability. It needs a rapid rehire authority targeted at the dispersed federal cohort before the window closes. It needs procurement reform that breaks the SETA contractor stranglehold so that any future transition program is not immediately captured by the same firms that benefited from the 2025 contractor windfall. It needs sustained funding for the technical career ladder inside government — competitive GS-14 and GS-15 pay for AI and data infrastructure roles, expanded direct hire authority, term-limited mid-career appointments.

None of this is on the political radar because the political radar is calibrated to form-layer activity. Capacity recovery is encounter-layer work and it does not produce press releases.

What the labs could actually do

The Sun piece describes the Anthropic Institute and Anthropic's $20 million to the Bores PAC as the leading edge of lab policy engagement. The OpenAI white paper is the most ambitious public articulation. Both are form-layer activity.

If the labs took their own diagnostic seriously at the encounter layer, the policy agenda would be different. They would fund the rebuild of federal technical capacity directly, not because it benefits them commercially but because their stated theory of change requires a functioning state to absorb the displacement they are causing. They would underwrite a 10x revival at $50 million a year and let it operate independently. They would pay for the rapid rehire program for the dispersed federal cohort before the window closes. They would lobby specifically and publicly for the boring procurement reforms that determine whether a transition program can actually run. They would treat the acquisition workforce vacancy rate as a leading indicator of their own reputational exposure, because a transition that fails operationally is the populist firebombing scenario Alex Karp warned about on the panel with Sean O'Brien.

The labs will not do any of this voluntarily. The transaction boundary is drawn at the white paper for a reason. Acknowledging the encounter-layer obligation means accepting that the residual public obligation is theirs to underwrite, which transforms the cost structure of frontier AI development in ways that materially affect their valuations. The white papers stay at the form layer because the form layer is where the boundary closes most cheaply.

This is what compelled redistribution looks like in policy form. The labs have to be made to underwrite the capacity rebuild as a condition of operating. Not as charity. As Vr settlement. The mechanism is some combination of windfall taxes, mandatory contributions to a transition fund, equity stakes in a public wealth vehicle that finances delivery infrastructure, and direct procurement set-asides that force the labs' enterprise revenue back into federal capacity recovery. The specific design is a policy question. The principle is that the entity capturing Vp has to underwrite the institutional capacity that will absorb the Ve and the Vr. Otherwise the partition stays where it currently sits, and the white papers become the historical artifact of a transition that nobody operated.

The piece nobody is writing

Sun's piece is the right diagnosis at the wrong level. It treats AI labor displacement as primarily a political and economic problem when it is also and more fundamentally a delivery capacity problem. The political and economic dimensions get most of the oxygen because they are familiar, contestable in op-ed length, and amenable to advocacy framing. The delivery capacity dimension gets none of the oxygen because it is technical, boring, and does not have a constituency.

The constituency that should be pushing this is the cohort that was just purged from federal service in 2025 — the people who actually know how transition programs run because they ran them, the people who know what 10x looks like at scale because they built it, the people who know what procurement reform requires because they lived through procurement failure. That cohort is currently dispersing into the private sector and losing the standing to make this argument with every passing month.

The transition has no operator. The capacity to operate it is being deliberately wasted in real time. The political class has not noticed because the political class measures progress at the form layer. The labs have not engaged with it because their boundary closes at the white paper. The civic tech foundations have not financed it because micrograms do not produce press releases.

The piece nobody is writing is the one that names the operational gap as the binding constraint and treats capacity recovery as the prerequisite to every other policy proposal currently in circulation. The displacement curve is real. The political opening is real. The federal apparatus required to convert the political opening into actual transition outcomes is being disassembled at speed. Somebody could pick up the work. Almost nobody is.

Sources

• Jasmine Sun, Silicon Valley Is Bracing for a Permanent Underclass, New York Times, April 30, 2026
• Nextgov/FCW, Government pacing toward increased IT contract spending despite DOGE cuts
• NPR, Federal agencies are rehiring workers and spending more after DOGE's push to cut
• Government Executive, Project 2025 wanted to hobble the federal workforce. DOGE has hastily done that, and more
• OpenAI, GDPVal benchmark
• Wired, Block layoffs and Jack Dorsey on AI restructuring
• Anthropic, research on AI coding agents and junior engineer skill development
• 10x.gsa.gov, What we do