The Agency Compact: Adopt Freely, Share the Upside

Doctrine in one sentence

Let firms and workers adopt useful AI, while making every resident a direct claimant on public value, a protected user of essential systems and a rights-bearing participant in transition.

Executive summary

The evidence supplied does not establish economy-wide wage collapse. UK AI use reached 29% of businesses in June 2026, and 49% among firms with at least 250 employees, but only 5% of AI-using businesses reported AI-related headcount reductions in March 2026. The ONS data are still developing, and DSIT says causal attribution is difficult. EU AI use was 20% among enterprises with at least ten employees in 2025.

That uncertainty is a reason to design institutions in advance, not to impose a blanket restraint on adoption. The package below is therefore a conditional settlement with five parts.

First, introduce a modest, taxable agency payment for adults, delivered through national tax and welfare systems. It would supplement, not replace, disability, unemployment and other targeted support. A protected capital account would sit alongside it, allowing every adult to hold a claim on a diversified public investment fund. The account would not be counted towards the United Kingdom’s £16,000 Universal Credit capital limit.

Secondly, finance the system from observable economic bases: broad profits, distributions, capital gains and economic rents, together with realised returns from public investment. There would be no attempt to calculate the percentage of a workflow performed by AI. No AI levy should be introduced until a reliable, difficult-to-avoid base exists.

Thirdly, give workers rights when employers introduce high-impact automated recruitment or management systems, or materially alter work through automation. The entitlement would be to notice, explanation, appeal, data access and portable transition support. It would not be a right to preserve commercially unnecessary tasks.

Fourthly, prevent the payment being capitalised into higher rents. Competition policy should improve portability and interoperability, while housing, energy and essential-service supply receive explicit scrutiny. Cash support without supply reform risks becoming a transfer to landlords, platforms and other holders of scarcity.

Finally, make expansion conditional on economy-wide outcomes rather than AI adoption itself. If real disposable incomes weaken while output or employment remains resilient, the payment can rise in pre-set increments. If the thesis is substantially wrong, the initial cost remains limited and the package retains ordinary benefits in poverty reduction, public investment, competition, worker due process and skills.

The central institutional division is deliberate. The UK and EU member states should provide recurring income through their own tax and transfer systems. EU institutions should coordinate rights, competition, procurement, cross-border portability and transition funding. The EU budget cannot run a deficit, direct-tax measures normally require unanimity, and monetary financing is prohibited. An EU-wide permanent income entitlement cannot responsibly be promised on the facts supplied.

The policy package

The package should be understood as a successor economic circuit, not as an attempt to repair the old one by preserving unnecessary employment. It links household purchasing power to public value and public investment rather than making every household dependent on a wage.

1. The agency payment. Each jurisdiction would establish an automatic payment for adult residents, administered through existing tax and welfare channels. It should begin at a deliberately modest level, be taxable or recoverable through the income tax system at high incomes, and be paid without an AI test or a work requirement. Existing targeted benefits should not be withdrawn merely because a person receives it.

The payment should be portable across employers, sectors and periods outside paid work. It could support care, study, entrepreneurship, reduced hours or a conventional job. Its purpose is not to replace work as a social good. Its purpose is to ensure that a person does not lose all economic agency when the market no longer needs the same amount of human labour.

2. The citizen capital account. Every adult would also receive a protected claim on a diversified public fund. The account would be individual and visible, but investment decisions would be pooled and professionally governed. People could receive cash dividends or reinvest them. The account should not be pledgeable as collateral and should not be forced into housing or other assets whose prices may rise in response.

This distinction matters. A cash payment gives immediate agency. A capital account gives a durable claim on future productivity. The fund would not need to own every successful firm. It could receive realised returns from commercial public investment, including equity or convertible investments where those are already legally and commercially appropriate.

The UK National Wealth Fund has £27.8 billion of capacity and a digital and technology remit. That does not automatically create citizen dividends. A new mandate, ring-fencing and governance rules would be needed before any part of its returns could be assigned to citizens. The full £27.8 billion should not be treated as annual spending money.

3. An adoption-compatible transition right. Workers affected by automated recruitment, worker management or other high-impact systems should receive notice, an explanation of the system’s role, a route to human review, access to relevant personal data and a right to challenge consequential decisions. The firm would also provide a transition plan where a material change in staffing or work organisation occurs.

The trigger is an observable management decision or system deployment, not an estimated percentage of AI in a workflow. A worker would not need to prove that a particular task was automated. Nor would the firm be prohibited from adopting the system. The right would be to due process and an exit route.

A portable transition credit should finance training, career changes, care periods, health-related retraining, or starting a small enterprise. It should be usable outside the original employer. Initial funding could come from existing adjustment, inclusion and training budgets, supplemented by national tax revenue. It must be evaluated before becoming a large permanent entitlement.

4. A supply and competition shield. The agency payment should be accompanied by a test of whether housing, energy and essential digital services can respond to additional purchasing power. Where supply is constrained, governments should prioritise measures that increase capacity, improve efficiency, capture publicly created land or infrastructure value, and protect consumers from lock-in.

Competition policy can require or encourage portability, interoperability and data access. This will reduce rents and dependence on dominant platforms, although it will not itself distribute income. Procurement can require transparency, portability, evaluation, skills and transition conditions where those conditions relate to the contract and are verifiable. It cannot be used as an unlimited route to unrelated equity claims or domestic preference.

5. A public return for public support. Where the state makes a commercial equity or convertible investment in AI infrastructure or firms, it should seek a financial return on commercial terms, with the possibility of assigning part of the net return to the citizen fund. Where support is a grant or procurement contract, conditions should stay within the legal limits of contract relevance, verification and non-discrimination.

This creates a reason for firms to accept adoption. They retain the gains from successful innovation, but society receives a transparent return when public capital is at risk. It does not require all countries to restrain adoption together. It works through national tax residence, market access, public support and public procurement.

United Kingdom: first 24 months

The first six months should be devoted to costing and institutional preparation. The Treasury should publish scenarios for a modest agency payment, the transition credit and the citizen fund. The OBR should score any permanent transfer or tax change. The modelling must include benefit interactions, tax recovery at higher incomes, administrative errors, labour supply, housing and energy pass-through, corporate avoidance, offshoring and distributional effects.

Parliament should then establish three rules: the payment formula, the protected status of the capital account, and the legal separation of citizen assets from general government spending. The account should be disregarded for Universal Credit capital purposes, because the present £16,000 rule could otherwise punish saving through the very institution intended to create security.

During months six to twelve, the government should appoint an independent trustee or equivalent steward for the citizen fund, with audited accounts, capped administrative charges and a diversified investment mandate. The National Wealth Fund should not be ordered to chase speculative AI assets. It should assess whether existing or future equity and convertible investments can generate commercial returns, and whether a citizen ownership sleeve is legally and financially sound.

The government should also establish a cross-sector baseline for workplace automated decision-making, using existing sectoral regulators and employment arrangements. The UK has no general AI Act or single AI regulator, so the first version should be narrow and enforceable: notice, explanation, human appeal, records, accountability and non-discrimination in high-impact workplace decisions.

During months twelve to twenty-four, the UK should launch the account and a modest payment, subject to the OBR score and a published funding source. It should pilot portable transition credits and contract-linked procurement conditions. The launch should include a public register of supported systems and a clear appeal route, without requiring the state to measure the exact AI content of each workflow.

No permanent payment should be funded by assuming future growth, unearned investment returns or an untested AI levy.

European Union and member states: first 24 months

EU institutions should use the AI Act, competition frameworks, procurement rules and existing adjustment and social funds to establish a common floor of process rights. Workplace recruitment and worker-management uses can be high-risk under the AI Act. Implementation should therefore focus on documentation, accountability, human review and worker access to remedies. The proposed later high-risk deadlines were not finally enacted on 12 July 2026, so the package must not depend on them.

The Commission and member states should develop common templates for worker notification, system evaluation, portability and appeals. These would reduce compliance costs for firms operating across borders without pretending that the EU budget can fund a permanent income entitlement.

EU institutions should also make public AI support more transparent. The EU plans to mobilise €200 billion for AI investment, including €20 billion for up to five gigafactories. That is a mobilisation figure, not a permanent dividend fund. Where public support involves investment, the public return should be recorded. Where it involves procurement, conditions should remain contract-linked, verifiable and non-discriminatory. Selective national support must continue to comply with State aid rules.

Member states should take the faster route on recurring income. Each government should establish a national agency payment, a protected capital account and portable transition support through its own tax and welfare systems. The amount and tax mix can differ, but the core principles should be interoperable: no AI-share test, no means-test penalty for protected public capital, and no dependence on continuous employment.

Member states should also publish housing and energy pass-through assessments before scaling payments. Existing EU adjustment and social funds can help with retraining and inclusion during the transition, but they should not be presented as a permanent EU income source.

Years 3 to 5 and dormant triggers

Years three to five should be a controlled expansion phase. The initial payment should not rise automatically merely because enterprise AI use increases. Adoption data are useful context, but entitlement should respond to household security, demand and public revenue.

Before year three, each jurisdiction should publish a baseline for real median disposable income, employment, underemployment, household housing and energy costs, payroll income, non-labour income, business formation and tax receipts. The baseline should be fixed before the programme begins and revised only through an independent review.

A scaling trigger should activate if, for two consecutive quarters, real median disposable income is at least 2% below its baseline while output or employment is stable or rising. A second trigger should activate if housing and energy costs grow faster than disposable income for four consecutive quarters. The first trigger would support a measured rise in the cash payment. The second would require supply and competition action before simply increasing cash.

A dormant capital trigger should apply when public investments generate realised net returns above their modelled capital and administration costs. Those returns would first replenish the fund and then support dividends. They should not be spent before they are realised.

By year five, the payment should be reviewed against three conditions: whether it reaches households reliably, whether it has been absorbed by rents, and whether the public finances can sustain it without worsening the approved debt path. If real incomes, employment and demand remain resilient for three years, temporary payment increases should pause. The protected account, due-process rights and competition measures should remain.

Funding and fiscal arithmetic

The proposal has two different funding problems, and they should not be blurred.

The annual cash requirement is:

eligible adults multiplied by the payment, minus tax recovery, plus administration and any benefit interaction costs.

The capital account is:

contributions plus realised net investment returns, minus dividends, fees and losses.

The supplied facts do not include the UK adult population, household income distribution, existing benefit expenditure, tax receipts, administrative costs or an expected fund return. It is therefore not honest to state a national pound cost for a universal payment. Those figures must be modelled before enactment.

The credible funding sources are:

• broad taxation of high incomes, profits, distributions, capital gains and economic rents, rather than a tax on the use of AI itself;
• financial returns from public equity or convertible investments made on commercial terms;
• reviews of tax expenditures and public support where the distributional case is strong;
• contract-linked public procurement conditions that reduce future transition costs, without pretending that procurement is a general ownership mechanism;
• existing adjustment and inclusion funds for temporary transition support.

The main payers would be higher-income households, owners of profitable firms and recipients of economic rents. Ordinary workers should not be charged more merely because their employer adopts software. Consumers should not be expected to finance the system through higher rents or platform prices.

The UK’s 15% employer National Insurance rate above the 2026/27 threshold is an observable base, but increasing it is not automatically desirable. A payroll-only approach could penalise labour-intensive firms and encourage avoidance. It should be considered only alongside broader profit and rent bases.

The UK National Wealth Fund’s £27.8 billion capacity and the EU’s €200 billion AI mobilisation are useful scale anchors, but neither is an annual income pot. The EU’s €20 billion gigafactory figure is included within the wider mobilisation and must not be counted twice. No borrowing from the ECB or national central banks should be assumed. Article 123 TFEU prevents monetary financing, and the ordinary EU budget cannot run a deficit.

The first two years should fund administration, legal preparation, auditing, pilots and a modest payment from voted national budgets. The EU level should use existing funds only for eligible transition and inclusion purposes. Every permanent commitment requires a published fiscal score, sensitivity analysis and a funding plan.

Political coalition and public case

The coalition is broader than a simple pro- or anti-technology alliance. It can include workers and trade unions seeking due process, firms seeking permission to adopt without indefinite uncertainty, households seeking security, technology companies seeking interoperable markets, public investors seeking a return, and fiscally cautious voters who prefer a small tested payment to an emergency response later.

The losers are clearer. High-income taxpayers and holders of economic rents would contribute more. Firms relying on opaque automated decisions would face new obligations. Some public managers would lose discretion over funds. Governments would have less freedom to raid a citizen fund for short-term budgets.

Compensation should be principled rather than universal exemption. There should be no retrospective confiscation, small firms should receive proportionate compliance requirements, and affected workers should receive portable transition support. Public investors should be paid according to commercial risk, not forced into politically selected assets.

The ordinary-language case is simple: if technology creates more value with fewer people, everyone should have a floor, everyone should share in public investment, and nobody should lose the ability to challenge a consequential machine decision. Firms remain free to innovate. Citizens receive a claim on the prosperity that public institutions help make possible.

Durability and anti-capture design

The citizen account should be governed by statute, with individual entitlements recorded separately from general government cash. Future governments should face a formal legislative process, published reasons and independent financial reporting before changing the formula or using the assets. This cannot make repeal impossible, but it raises the political and administrative cost of a raid.

Investment should be diversified, passive where appropriate and subject to clear conflict-of-interest rules. Ministers should not choose individual companies for citizen ownership. The trustee should publish holdings, returns, fees, voting policies and realised losses. A public audit body should inspect both the fund and the payment system.

The payment formula should be automatic enough to limit annual bargaining, but Parliament should retain responsibility for its level and tax treatment. Appeals should be available for missed or miscalculated payments. No private technology provider should be the sole administrator of the citizen ledger.

Avoidance controls should attach to consolidated economic activity, distributions, public support and market access. Any future compute, robot, token or AI levy would need a defined base, import treatment, anti-avoidance rules and a clear interaction with existing taxes. If those conditions cannot be met, the levy should not proceed.

Offshoring risk means the package should not rely on one narrow domestic tax. Member states should coordinate where possible, while each country retains a fallback based on its own broad tax system. Public support should be conditional on verifiable contract or investment outcomes, not nationality.

Legal and institutional obstacles

The UK can alter tax and welfare parameters through Parliament, but permanent payments need funding and OBR scoring. The Universal Credit capital rule would need adjustment for protected accounts. The absence of a general AI Act and single regulator makes cross-sector consistency difficult. A narrow baseline using existing sectoral institutions is more feasible than an immediate comprehensive regulator.

The National Wealth Fund can use equity and convertibles, but its present mandate does not automatically establish citizen dividends. Any citizen ownership arrangement requires legal, governance and financial work. Procurement conditions must relate to the contract, be verifiable and non-discriminatory. They cannot be used to require unrelated parent-company equity.

At EU level, direct taxes and new own resources normally require unanimity. The ordinary EU budget cannot run a deficit, and Article 123 TFEU blocks monetary financing. The permanent income layer must therefore be national. EU institutions can coordinate standards, competition, procurement and transition funding, but should not imply that existing social funds already provide an income entitlement.

Member-state support for AI remains subject to State aid rules. National governments must also reconcile workplace rights with the AI Act and their own labour and social-security systems. Before enactment, legal reviews should test tax incidence, cross-border avoidance, public investment rules, procurement relevance, State aid compatibility and the treatment of mobile firms.

Failure modes, review and exit rules

If the thesis is wrong, a large permanent payment could become an expensive transfer with little connection to need. The bounded-regret response is a modest initial amount, automatic tax recovery at high incomes, staged expansion and a continuing fiscal review. Anti-poverty, competition, due process and public-return measures would remain useful.

If the thesis is right but the payment is captured by landlords, energy providers or platforms, the rent trigger should require supply and competition action. The government should publish pass-through evidence rather than simply raising the payment.

If firms reclassify activity or move it abroad, the response should be to broaden the observable base and improve enforcement, not to estimate AI content more aggressively. If an independent audit finds that more than 15% of expected contributions are being lost to avoidable leakage for two consecutive years, expansion should pause until the base is repaired.

If the fund underperforms, dividends should fall or pause, but the cash floor should not depend on speculative returns. If administrative error exceeds a published tolerance, new enrolment should continue only with a correction plan and an accessible appeals process.

The principal review points should be month twelve, month twenty-four and the end of year five. Scaling down means stopping temporary increases, not removing established rights or confiscating earned account balances. Ending a plank should require evidence that it is ineffective, regressive or fiscally unsafe. No exit rule should allow a government to raid the capital account merely because a deficit is inconvenient.

The indicators should be published in a single dashboard: real median disposable income, employment and underemployment, housing and energy costs, payment reliability, public investment returns, tax leakage, platform concentration and worker appeals. AI adoption data can be shown beside them, but it should not be the decisive variable.

Feasibility table

The ratings use the following test. High means the institution already has a clear competence and could begin within two years if funded. Medium means it is legally possible but needs new legislation, significant coordination or difficult fiscal choices. Low means the action conflicts with the stated institutional constraints or would require agreement that is unlikely to arrive quickly.

What is genuinely new here

The distinctive element is not a new claim about AI. It is the separation of three questions that are usually combined.

The first is permission to adopt. Firms are not asked to preserve tasks that no longer make commercial sense. The second is the public claim on adoption. Where public capital, public procurement or publicly created market conditions help produce value, citizens receive a transparent route to returns. The third is household agency. A person receives cash security, an asset claim and a route through consequential workplace decisions.

The second innovation is institutional rather than fiscal. It accepts that recurring income will be national, while the EU concentrates on rights, competition, procurement and coordination. That fits the supplied legal constraints better than promising a central European dividend.

The third is the trigger design. Policy expansion responds to real disposable income, demand, costs and fiscal capacity, not to a contested measure of how much AI is inside a workflow. The thesis can therefore be tested without making the success of the policy depend on proving the thesis in advance.

Bottom line

The safest response is neither to defend every existing job nor to wait for unmistakable wage collapse. Establish a modest national agency floor, a protected citizen claim on public investment, adoption-compatible workplace rights and supply-side protection against rent capture. Let the evidence determine whether the floor grows. This preserves economic agency if the thesis is right, while retaining ordinary public value if it is wrong.