Issue #50

Block Cuts 4,000 Jobs, Reviving the AI Tax Debate

Jack Dorsey's layoffs and Andrew Yang's push to tax AI raise a hard question: who pays when machines do the work?

Society

Oswarld (Kwangseob Ahn)· Writer & editor of INLEVEL9

Add as a Google preferred source
Block Cuts 4,000 Jobs, Reviving the AI Tax Debate

Block’s Layoff Announcement and the AI Tax Debate

On February 26, 2026, Block, the fintech company led by Jack Dorsey, announced a plan to cut more than 4,000 jobs. That’s a reduction from over 10,000 employees to fewer than 6,000. In his letter to shareholders, Dorsey explained that AI tools now let smaller teams accomplish more work. Reports also noted that the stock jumped more than 24% in after-hours trading following the announcement.

Among American politicians, I’ve been paying close attention to Andrew Yang, who is of Taiwanese descent. He’s long argued for a universal basic income. In a CNBC interview in March, he proposed cutting taxes on labor and taxing AI instead. Reading that alongside Block’s layoff announcement, I found myself wondering: as companies become more productive, how do we guarantee a living for the people who lose their jobs in the process?

Why We Need to Look at Hiring Slowdowns, Not Just Layoffs

When examining employment shifts, we need to watch new hiring as closely as layoffs. Even without headline-grabbing mass layoffs, job opportunities can shrink simply because open positions go unfilled. In a May 2025 interview, Anthropic CEO Dario Amodei warned that half of entry-level white-collar jobs could disappear within one to five years, potentially pushing unemployment to 10-20%. That’s his forecast, not a finding that this has already happened.

According to New York Federal Reserve data, the unemployment rate for 22-27-year-old college graduates stood at roughly 5.7% in Q4 2025, with an underemployment rate of 42.5%. Here, underemployment1 refers to working in a job that typically doesn’t require a degree. These figures illustrate the difficulty of early-career job searches, but shouldn’t be read as the overall unemployment rate for college graduates across all ages.

An analysis Goldman Sachs presented in February 2026 found that industries with a high share of college graduates saw average monthly employment change of -9,000 from 2023 to 2025, while low-share industries saw +12,000. However, this analysis didn’t attribute the deterioration in the college-graduate labor market at the time to widespread AI impact. It noted that both hiring and job-switching had slowed broadly, and discussed AI’s future risks as a separate matter.

I think that for job seekers, shrinking job postings matter just as much as news of mass layoffs. When assessing AI’s impact, we also need to distinguish between the situation facing people already employed and those just entering the workforce.

If Wages Shrink, Where Do Taxes Come From

US federal tax revenue relies heavily on individual income tax and payroll taxes for things like social security. Individual income tax covers not just wages but also business and investment income. So lumping the two together and calling it a “tax on labor” isn’t quite accurate.

If AI adoption shrinks total wages, the tax base tied to wages and payroll taxes could weaken. But if corporate profits and investment income rise instead, how that income gets taxed becomes the crucial question. Even if productivity climbs, employment, profits, and tax revenue don’t necessarily grow at the same rate.

Yang’s proposal aims to lower the tax burden on hiring people, and instead fund things from wherever AI generates economic value. But simply saying “tax AI” doesn’t settle who pays or what the tax base is. You still have to decide whether the target is the profits of companies that build AI, usage fees, or automated services themselves.

Amodei floated the idea of a “token tax”—collecting some share, say 3%, of the revenue AI companies earn from model usage and redistributing it. This isn’t a concrete bill or a fixed rate. And despite the name “token,” the example he gave leans closer to applying a percentage to usage revenue than charging a flat fee per token.

Different Approaches to What Gets Taxed

When I read through AI tax proposals, I sort them into three buckets: taxes on profit, taxes on usage fees, and taxes on automated work.

Even when the same tax rate is proposed, how much money gets collected — and who actually bears the burden — depends on what the tax applies to.

The first approach taxes corporate profit and capital income. A 2024 IMF report addresses strengthening capital income taxation and corporate tax enforcement, and revisiting tax incentives that encourage excessive labor displacement. It also raises the possibility of taxing monopolistic excess profits generated by AI. This isn’t a proposal limited to a separate tax category just for AI companies.

The second approach taxes AI service fees or usage volume. A January 2026 Brookings piece by Anton Korinek and Lee Lockwood treats AI services delivered to end consumers as a target for consumption taxation. At the same time, they point out that collecting taxes at every stage of business-to-business transactions could create a cascading burden problem2.

The third approach taxes based on automated work itself. Jack Kidd, founder of AskHumans, proposes a “task tax” levied on individual jobs performed by robots. His hotel-cleaning example assumes replacing $28-an-hour human labor with a $2 robot, and using part of the savings as a funding source. This figure is an illustrative assumption used to explain the proposal.

Each approach has its own difficulties. Taxing profit requires figuring out where and how much profit was generated. Taxing usage fees requires distinguishing between final consumption and a business’s production activity. A task tax requires deciding which portion of work done jointly by AI and humans counts as “automated.” That’s why the standard for what gets taxed needs to be settled before the tax rate does.

Is There Enough Money to Fund a Universal Basic Income

How you collect taxes and how you spend what’s collected need to be discussed together. The universal basic income that Yang has long championed is one possible use for that revenue.

For his 2020 presidential campaign, Yang proposed a “Freedom Dividend” giving every US citizen aged 18 and over $1,000 a month. His funding plan included a 10% value-added tax along with other tax increases. In a 2019 modeling analysis, the Tax Foundation estimated that even with that combination, funding would fall short—and that the average marginal tax rate on labor income would rise by about 8.6 percentage points, potentially shrinking long-run output by roughly 3%. This estimate reflects the assumptions of that specific proposal and model at the time.

If AI raises productivity, it’s possible that new sources of revenue could emerge. But not all the money a company saves on labor costs becomes taxable additional profit—AI usage fees, capital investment, and transition costs all eat into it. To have a real conversation about basic income, projected tax revenue and the funds needed for payouts have to be calculated on the same basis.

A 2025 IMF study on Korea found that roughly half of all jobs are exposed to AI’s effects, with women, young people, and highly educated workers showing higher exposure. “Exposure” here covers both jobs being replaced and jobs becoming more productive with AI’s help.

A 2023 report by Han Yo-seop, a research fellow at KDI (Korea Development Institute), estimates that the short-term impact on overall corporate employment is small, but regional analysis shows negative effects on employment and wages for some segments of younger workers. The report also cites survey findings that companies planning to adopt AI expect to cut new hiring more than they expect to reduce their existing workforce.

President Lee Jae-myung also referenced an “AI Basic Society”—one where the benefits of technological progress are shared broadly—at the 2025 APEC summit. Turning this vision into an actual policy will require deciding how to fund it and who gets what kind of support.

Oswarld’s Lens

I think what matters more than the name “agent tax” is who actually captures the gains from higher productivity. Investors may read Block’s layoff announcement as a sign of improved profitability, but the people losing their jobs have to worry about where their next paycheck comes from. Pointing to a higher company valuation doesn’t resolve the losses individuals actually experience.

The research I’ve read doesn’t converge on a single fix. IMF researchers lean toward strengthening taxes on capital income, while Brookings’ Kearney and Lockwood emphasize taxing final consumption while preserving investment incentives. I think we need to hold this divergence in view and examine where the current tax system falls short.

My first question is whether we can support the livelihoods of people who lose their jobs or face longer job searches. A system that has funded itself through wage-based revenue needs to hold up even as employment patterns shift. We also need to check whether existing corporate and income tax structures are missing certain gains, and whether the benefits flowing to automation are badly out of balance with support for employment.

If opportunities shrink for people just finishing college and trying to start working, the expectation that work can sustain a livelihood and build a career will weaken too. That’s why I find Yang’s line of questioning worth paying attention to. The American experience can’t be mapped onto Korea one-to-one, but I think the first jobs and career formation of young Koreans deserve to be examined with the same seriousness.

Closing

When designing AI taxation, we need to check who actually pays the tax, who ends up bearing the real burden, and where the money collected gets spent. I do believe we need some system to share the benefits of productivity gains, but I don’t think a single tax name or a low tax rate alone counts as an adequate solution.

We need to prepare two separate tracks of policy: one that addresses the employment and tax-revenue shifts happening now, and another for a future where far more autonomous AI is directly engaged in economic activity. For now, we can start with something concrete — tracking changes in hiring, wages, and corporate profits to calculate exactly who needs support and where the funding should come from.

Your take shapes the next issue

What resonated most in this issue, or where has your experience been different?

Any registered reader can comment for free.

References & Further Reading

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

Footnotes

  1. Underemployment: in the New York Fed statistics cited in this piece, this refers to college graduates working in jobs that typically don’t require a degree. It is not a catch-all indicator for job dissatisfaction or non-regular employment. ↩

  2. Tax cascading: a problem where tax is applied at every intermediate transaction without input tax credits, so tax compounds into the cost of the next stage. Value-added tax is designed to reduce this cascading through input tax credits. ↩

Related issues

Issue #45

Virtual Cards Built Just for AI Agents

How much spending power should an AI agent get? Prepaid limits, approvals, and dispute handling are the open questions.

Society
Issue #38

Why Chinese Firms Are Racing to Install OpenClaw

From install lines to messenger hookups, here's what my Chinese clients are telling me about OpenClaw's sudden rise.

Society
Issue #37

How Yakult Delivery Workers Check on Their Neighbors

Japan's Yakult Ladies and Korea's hy delivery workers quietly track elderly customers' well-being during routine drop-offs.

Society