Since launching the Anthropic Economic Index, Anthropic has observed a notable behavioral shift among users: instead of collaborating step-by-step with AI systems, many now delegate complete tasks to Claude. The company expects this pattern to accelerate as more employers adopt AI to improve productivity and reduce operational friction.

That evolution signals the beginning of a broader transition, one that many workers and leaders have been sensing for some time. If AI increasingly handles complex workflows end-to-end, it could reshape how labor, wages, productivity, and economic value are distributed.

With meaningful uncertainty ahead, Anthropic says these early patterns underscore the importance of proactive planning. As capabilities advance, the company argues it is crucial to begin open discussion about the tools policymakers may need to respond to AI’s economic effects, whatever form they take. To support that effort, Anthropic is releasing a set of economic policy ideas that it believes merit further study.

Anthropic notes that predicting AI’s economic impact remains challenging. Leading economists disagree on whether AI will:

That uncertainty, the company argues, does not diminish the need for early planning. Instead, it increases it.

Historically, major economic transformations — from industrial automation to the globalization wave — have shown that policy responses often arrive late, once job losses or wage pressures are already underway. Anthropic says it hopes to avoid repeating that pattern by surfacing ideas while the window for proactive policymaking remains open.

Over the past year, the company engaged economists, academics, and policymakers — including participants from its Economic Advisory Council and its Economic Futures Symposium — to explore tools that could help governments adapt. The effort intentionally included experts from across ideological perspectives, including non-partisan contributors, to ensure debate reflects a wide range of views and isn’t guided by a single political lens. The goal is to widen public discussion and make early research transparent, rather than advocate for specific policy outcomes.

As Anthropic puts it, planning now helps ensure workers and public institutions are not caught off guard when AI adoption accelerates.

The rate, scale, and shape of AI’s economic effects will determine which policy tools governments may ultimately need. Anthropic notes that no single path is guaranteed: AI could boost productivity with modest labor disruption, or it could reshape employment and public finance far more dramatically.

To help governments think ahead, the company organized ideas into three broad scenario categories based on potential labor-market impact.

Even in scenarios where AI’s impact on jobs remains limited and largely productivity-driven, experts say governments will still need to strengthen workforce systems and accelerate infrastructure capacity to support innovation and growth. Some of these ideas are not new, but are gaining renewed relevance as AI advances.

These proposals are designed to help workers and businesses adapt to evolving roles and skill demands.

At Anthropic’s Washington D.C. Economic Symposium, Abigail Ball, Executive Director of American Compass, highlighted a “Workforce Training Grant” model developed with Oren Cass. The proposal would fund formal trainee positions and structured training programs directly through employers.

Governments would provide substantial annual subsidies to employers (Ball and Cass suggest $10,000 per trainee in the United States) to create formal AI trainee positions and training programs.

Training could be delivered through individual companies, coalitions of employers, unions, or partnerships with community colleges and technical schools.

American Compass suggests funding the program by redirecting existing higher-education subsidies. However, the organization notes that other funding approaches may also merit consideration, including potential taxes on AI consumption to help support workforce-development efforts.

The goal is to support hands-on skill development for learners entering new fields and workers transitioning within evolving industries.

Revana Sharfuddin of the Mercatus Center argues that today’s tax code favors capital investment over human capital. Businesses can immediately expense AI systems through bonus depreciation, but face tighter limits and reporting requirements when deducting worker-training expenses.

Her proposal calls for reforms to the Internal Revenue Code, including:

These reforms aim to make retraining more financially attractive than layoffs, helping support workers whose jobs sit at the margin of automation risk.

Tax policy expert David Gamage has outlined proposals to modernize tax structures to ensure governments capture value from digital-first and AI-enabled businesses.

His suggestions include:

The goal is to reduce artificial profit-shifting into tax havens, a practice that could become more common as AI increases the share of economic value tied to intangible assets.

According to Gamage, “governments that act first will solve their fiscal challenges and better position residents to thrive in an AI economy. Those that wait will face resource constraints when flexibility is most needed."

Anthropic stresses the urgent need to modernize U.S. permitting systems and power procurement processes to scale:

Today, environmental reviews, state regulatory processes, and grid interconnection approvals can stretch 4–10 years or more — delaying AI infrastructure needed to support rapid AI adoption.

Concrete reforms under discussion include:

Faster approvals could help attract investment, support economic growth, and create jobs in regions where AI infrastructure is developed. Without reforms, delays may slow productivity gains, limit domestic capacity, and increase the risk that critical AI infrastructure is built overseas, raising longer-term competitiveness and national-security concerns.

Tyler Cowen, faculty director at the Mercatus Center and member of Anthropic’s Economic Advisory Council, supports this view, saying, “I am all for permitting reform — the energy sector included.”

As AI adoption accelerates and automation affects a broader share of roles, economists are exploring targeted models to help displaced workers transition into new fields and maintain income stability. Below are key policy proposals under consideration.

Economists are examining how the federal Trade Adjustment Assistance (TAA) program — which provides retraining, education benefits, job-search assistance, and income support for workers displaced by economic shifts — could serve as a template for helping workers adapt to AI-driven changes in the labor market.

Ioana Marinescu, University of Pennsylvania economist and member of Anthropic’s Economic Advisory Council, proposes a TAA-inspired “AI insurance” system to:

Researchers Suchet Mittal and Sam Manning suggest a similar model called Automation Adjustment Assistance (AAA). Key features include:

The goal across these proposals is to give workers direct support, time, and resources to re-skill as labor markets evolve, ensuring that productivity gains from AI translate into shared opportunity rather than concentrated disruption.

If AI adoption accelerates and automation begins to reshape economic activity more broadly, some economists believe targeted taxation tools may be necessary to help fund transition support programs and maintain fiscal stability.

University of Virginia economists Lee Lockwood and Anton Korinek (members of Anthropic’s Economic Advisory Council) are examining how governments might tax economic activity driven by advanced AI. Their research explores options such as taxing AI-generated tokens, digital services, or automated “robotic labor” in cases where AI systems displace human workers.

As they explain, tax approaches may need to evolve alongside AI adoption. In the near term — while humans still make most economic decisions — one model under consideration is a “token tax” on AI-generated outputs sold to users. This approach would allow governments to capture a share of AI-generated value without slowing early innovation or productivity gains.

However, they note that this model would be effective only while humans remain central to economic activity. If AI systems take on a much larger role in production and consumption, policymakers may need alternative approaches to avoid discouraging investment or slowing long-term growth.

If the economy reaches a point where advanced AI systems not only perform a large share of work but also consume significant compute and energy resources, Korinek and Lockwood argue that policymakers may need to shift from taxing AI-generated tokens sold to users toward taxing the infrastructure that powers AI.

In this scenario, taxes on compute, specialized hardware, or high-capacity AI systems could be more effective than taxing human users, since AI, rather than people, would be driving much of economic activity.

Under this model, human labor’s share of the economy would decline, making traditional wage- and consumption-based taxation less viable. Taxing AI infrastructure would provide governments a way to ensure AI-driven gains continue to support public services and social programs.

Here, the policy focus shifts from protecting innovation to ensuring that automation-driven gains continue to fund public needs and shared prosperity.

Anthropic notes that these proposals would shape AI companies’ economics, including its own, but says they warrant study given how government revenue systems and employment patterns may change as AI advances.

In a faster, more disruptive scenario, advanced AI could reshape labor markets rapidly, concentrating economic gains while creating sharper inequality and accelerated job displacement. In such a case, a small number of firms with access to compute, talent, and data could capture a disproportionate share of economic value, while a broad range of workers — from customer support and administrative staff to software and data professionals — experience faster-than-expected disruption. Without sufficient support mechanisms, those shifts could widen economic divides as automation scales.

In response, policymakers may pursue more ambitious measures, including models that give citizens a direct stake in AI-driven gains. Governments could also explore new revenue systems to sustain core social and economic programs in a future where traditional labor-based taxes may no longer be sufficient.

Some economists and policy researchers argue that if AI companies capture a disproportionately large share of national wealth, governments may need new mechanisms to ensure citizens benefit from future productivity gains.

One idea involves governments taking partial ownership stakes in major AI companies or core AI infrastructure, allowing the public to share in long-term economic gains rather than relying primarily on wage-based taxes. A comparable model exists in Alaska, where the state invests a portion of oil revenues into a public permanent fund and distributes annual dividends to eligible residents.

For example, Alaska announced a $1,702 dividend for 2024, and payments historically range from roughly $1,000 to $3,000 per person depending on fund performance and state budget decisions. Some economists suggest that a similar “AI fund” model could ensure citizens directly benefit if AI becomes a primary source of national wealth.

A similar concept from the U.K., put forward by Emma Casey, Emma Rockall, and Helena Roy, envisions an “AI Bond” program: government-backed investment in the national AI stack, with returns distributed broadly across the population. Proponents say these approaches could help maintain economic stability and public trust if wealth becomes more concentrated in advanced AI platforms and ecosystems.

Another idea under consideration is the use of value-added taxes (VATs). A VAT is a type of consumption tax, meaning it’s based on what people buy rather than what they earn. In simple terms, it works like a sales tax — but instead of only being charged at checkout, it’s collected gradually throughout the production process.

For example, a VAT can be applied when raw materials are purchased, again when parts are manufactured, and again when the final product is sold to consumers. This system already exists in places like the European Union, United Kingdom, and Canada. In fact, 6 of the G7 countries and 37 of 38 OECD nations already use VAT systems — the United States is the only exception in both groups.

Supporters say that if wages make up a smaller share of the economy as AI advances, shifting toward a consumption-based tax system could help ensure governments maintain a stable revenue base to support public services and social programs.

If AI ends up generating a larger share of economic value — reducing the portion from human labor — governments may need new revenue sources to supplement traditional income-based taxes.

David Gamage has outlined a hybrid approach combining traditional income taxes with a low-rate business wealth tax. The idea is to ensure companies generating large profits from AI continue contributing to the tax base, even if labor income no longer reflects the economy’s primary value driver. Gamage notes that income taxes alone can be subject to accounting strategies that minimize liability, while wealth taxes face valuation challenges; pairing both could reduce avoidance and better align the tax system with an economy where corporate profits may grow faster than wage income.

Gamage compares this idea to how investment managers charge fees. In his analogy, a business wealth tax would function like a “management fee” paid in exchange for the legal, financial, and economic frameworks provided by the public sector that allow companies to operate and scale. The income tax would then act like a “performance fee,” charged when firms generate profits.

In simpler terms, the proposal suggests that companies benefiting most from AI-driven growth would help fund the public systems that support their operations, especially in a future where human labor may represent a smaller share of economic activity.

Anthropic notes that the ideas do not reflect its official policy positions:

"The proposals below don’t necessarily represent Anthropic's own policy positions. But we’re excited by the breadth of proposals we’ve received, and we hope they encourage further research and debate."

The company emphasizes that these are not final recommendations and that collaboration with policymakers, researchers, and labor-market experts will remain essential as capabilities evolve and real-world data emerges.

Anthropic recently announced a $10 million expansion of its Economic Futures Program, aimed at deepening research into AI’s economic impacts and potential policy responses. The funding will support empirical research, the development of economic policy ideas, and global convenings, including a London symposium following an earlier event in Washington, D.C.

The company emphasizes that the ideas surfaced through this work are not policy endorsements, but proposals worth examining as AI adoption accelerates. Anthropic notes that the economic effects of AI remain uncertain — both in timing and scale — and that different adoption paths may require different responses.

At this stage, the priority is rigorous research and public discussion. Anthropic says that engaging economists, policymakers, and industry early can help ensure societies prepare thoughtfully for multiple possible futures, and that the benefits of AI are broadly shared.

Q: Why plan economic policy before impacts are clear?
A: Historically, major shifts have shown that responding late limits support options for workers and slows adaptation. Preparing early provides governments more tools and more flexibility.

Q: Does Anthropic endorse these proposals?
A: No. The company says these are concepts for study and debate, not policy positions.

Q: Why focus on multiple scenarios?
A: AI’s economic effects may vary significantly. Planning for low, medium, and high-disruption futures ensures flexibility.

Q: Why address permitting and infrastructure?
A: AI progress depends on energy, compute, and transmission capacity. Slow permitting could bottleneck innovation and national competitiveness.

Q: Who is advising this research?
A: A mix of academic economists, policy scholars, and industry advisors, including the company’s Economic Advisory Council.

Anthropic’s policy exploration reflects a pivotal moment in the AI era: we are at the start of AI-driven economic transformation, and institutions are preparing before impacts scale.

While no single blueprint exists, the company argues that early research, transparent debate, and scenario planning can help ensure workers, governments, and communities are equipped for multiple possible futures.

The question is no longer whether AI will influence the economy — it is how well society plans for the transition and who stands to benefit from it. Good preparation now could help ensure AI-era prosperity is widely shared, not concentrated. Without it, the adjustment could be harder and more uneven.

If AI delivers broad productivity gains like it says it will, planning may support faster growth and skill development. If disruption accelerates, early preparation may help soften shocks and ensure benefits are more widely shared.

This moment is not about sounding alarms.

It’s about recognizing an opportunity: to build the economic foundation of the AI age deliberately, responsibly, and in the open — while there is still time to shape it.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant used for research and drafting. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.