“I’m certain AI will do 80 per cent of the economically valuable work humans do today, for 80 per cent of all jobs, faster than most believe. The question isn’t whether mass underemployment arrives by the next decade, but whether we have a coherent policy framework ready when it does.” – Vinod Khosla – Experienced technology investor

Mass underemployment driven by automation is no longer a fringe scenario but a serious macroeconomic risk windowed into the next decade, not the next century.¹ The central tension is brutally simple: digital systems that learn are scaling faster than labour markets, education systems and tax architectures can adapt. Productivity, deflation and abundant digital services sit on one side; concentrated capital ownership, job displacement and fiscal stress sit on the other.^10,11 How governments, firms and citizens navigate this gap between technological tempo and institutional inertia will shape whether the transition feels like a ladder up or a trapdoor down.

From incremental automation to general task displacement

Traditional automation waves targeted specific routine tasks in manufacturing and clerical work. The current generation of AI systems differs in two structural ways. First, they are increasingly general-purpose: the same base model can be fine-tuned or prompted to perform tasks across domains, from drafting legal briefs to debugging code to producing marketing assets.^8,10 Second, their performance improves with more data and compute rather than bespoke engineering, which means capability gains propagate broadly and rapidly once new model families arrive.⁸

When observers argue that a very large share of economically valuable work in most occupations can be done by AI, they are pointing to the task composition of jobs rather than job titles.^1,10,13 A radiologist, an accountant and a sales executive each spend much of their day on information processing: synthesising documents, interpreting signals, generating options and crafting responses. These are precisely the activities modern large language models and associated tools are increasingly competent at automating or augmenting.⁸ The same model families can then, in principle, be deployed into hundreds of millions of roles, limited primarily by integration, regulation and organisational willingness rather than by domain-specific engineering.

This is why some investors now describe the near-future workplace as one in which AI systems perform the bulk of tasks in most jobs: not eliminating all human labour, but hollowing out enough of the task bundle that one human can supervise far more output or that many roles simply no longer justify a salary.^10,13 On this view, the labour market shock is not confined to a narrow band of routine manual work but extends directly into white-collar professions previously considered automation-resilient.^1,6,10

The timeline shock: why “faster than most believe” matters

Predictions around dates are always contestable, but the strategic issue is the gap between how quickly AI capability curves are steepening and how slowly labour-market institutions move. The claim that by around 2030 AI systems could be technically capable of doing most tasks in most jobs is not presented as a distant science-fiction scenario, but as a plausible extension of current trends in model scaling, multi-modal capabilities and robotics.^1,10,13 Several public interventions have suggested timeframes of roughly 5 to 10 years for AI to handle about 80\% of the task content in a similarly large share of occupations, with exceptions concentrated in hands-on and complex interventional domains such as heart or brain surgery.^12,13,15

Whether the figure is 60\%, 80\% or 90\% is less important than the compression of the adjustment window. Education systems, retraining pathways and social insurance mechanisms typically operate on decade-scale reform cycles. Corporate IT and process change move more quickly, especially once the economic case for adoption hardens around significant cost savings and competitive pressure.^7,13 The result is a structural timing mismatch: firms can move in years; governments habitually move in election cycles; individuals build careers over decades. A ten-year window in which AI becomes a general substitute for most labour tasks risks overwhelming slow-moving systems unless policy is pre-emptive rather than reactive.^8,11

Economic abundance built on labour displacement

The economic narrative accompanying these aggressive automation predictions is not one of collapse but of abundance. If AI and robotics drive the effective marginal cost of labour-intensive services towards zero, many goods and services could become dramatically cheaper.^10,15 Investors making these claims argue that by the 2040s the purchasing power of a moderate income could be an order of magnitude higher than today, with housing, education, healthcare and much routine consumption available at a fraction of current cost.^10,15

The mechanism is classically deflationary. Suppose an economy has output Y produced by a mix of capital K and labour L. In a stylised production function Y = F(K,L), AI-driven capital deepening effectively shifts productive capacity so that, for many sectors, \partial Y / \partial L shrinks towards zero while \partial Y / \partial K rises, because AI systems and robots stand in for human workers. If a large portion of L can be replaced by AI capital, then for the same wage bill total output can rise sharply, or for fixed output, labour requirements can collapse. Either way, the labour share of income falls while capital share increases.^10,15

The optimistic view emphasises that even if wages fall or jobs vanish, the required income to enjoy a high standard of living could fall faster. In that world, the binding constraint shifts from access to high wages to access to the abundance produced by AI. Proposals like universal basic income, sovereign AI wealth funds and near-free public services aim to recycle AI-driven returns on capital back to citizens.^10,15 The challenge is less technological feasibility than political economy: who owns the AI capital, how it is taxed, and how those taxes are redistributed.

Mass underemployment as a systemic risk

The claim that mass underemployment is not a distant possibility but a likely outcome within the next decade stems from the task-level analysis of jobs combined with the economics of AI deployment. Automation decisions are rarely made with macro employment in mind; they are made at the level of firm cost structure and competitive survival. Once AI systems can reliably handle most of the value-creating tasks in a role at significantly lower cost than a human employee, boards and executives face strong incentives to restructure staffing, often aggressively.^7,13

Underemployment risks emerge even if headline unemployment remains lower. Workers may retain some work but at fewer hours, weaker bargaining power and more precarious contracts. In sectors like call centres, back-office processing, basic coding, accounting and parts of legal and medical practice, there is credible evidence that large swathes of tasks are already partially automated with today’s models.^1,10,13 As deployment scales, firms can maintain service output while reducing human-hours demanded. The macro effect is a labour market in which jobs still exist on paper but become harder to access, more fragile and less well-paid compared with the historical productivity-sharing bargain.

Several institutional analyses, from international organisations to think tanks and AI labs, now explicitly consider scenarios of double-digit unemployment or large-scale labour displacement triggered by general-purpose AI.^8,11,14,17 Policy frameworks under discussion include scaled-up unemployment insurance, rapid retraining schemes, wage insurance, and various models of basic income or capital accounts that give citizens direct exposure to AI-driven equity returns.^8,11,17 The recurring theme is that relying on traditional slow-moving welfare systems and incremental labour regulation will be inadequate if the displacement wave arrives as quickly as some technologists project.

The tax architecture problem: where will revenue come from?

One of the most concrete tensions raised by these predictions is fiscal. Today’s tax systems in advanced economies are heavily reliant on labour-based revenue: income tax, payroll tax and consumption taxes funded by wage income. If AI significantly compresses the wage bill while lifting profits and capital gains, that base erodes. Yet the social demands on the state would simultaneously expand: income support, retraining, healthcare and housing assistance for those who struggle to find work in an AI-heavy economy.^1,10,11

Several proposals attempt to square this circle by shifting the tax base from labour to capital. One family of ideas focuses on taxing capital gains at similar or higher rates than ordinary income, especially for very high earners, on the grounds that AI-driven wealth will accrue disproportionately to holders of tech equity and intangible assets.^10,15 Another emphasises a national or regional sovereign wealth fund that accumulates stakes in AI and complementary technologies, using dividends and capital appreciation to fund social transfers and public goods.^10,15 A third explores explicit taxes on automated labour, sometimes framed as robot or AI usage taxes, though these raise difficult measurement and innovation-incentive questions.^8,11

The core mathematical intuition is straightforward. If aggregate labour income W stagnates or falls while capital income R rises sharply, and if government revenue T is currently something like T = \tau_W W + \tau_R R, maintaining or expanding public spending requires shifting the relative tax rates \tau_W and \tau_R. Without such adjustment, the tax base shrinks even as social demands rise. Designing that shift in a way that preserves investment incentives, avoids large-scale avoidance and remains politically legitimate is one of the defining policy design challenges of the AI era.

Work, purpose and the politics of not needing a job

Beneath the fiscal and macroeconomic arguments lies a more human question: what happens to societies built on the moral centrality of work when large numbers of people no longer need jobs to survive, or cannot find them? Some technologists argue that much modern employment is a form of economic servitude, and that freeing humans from the need to work could unleash a flowering of creativity, care and self-directed projects.^15,21 Others warn that work is a core source of identity, social connection and status; strip it away without robust replacement institutions and you risk alienation, polarisation and social unrest.

Political systems are not neutral in this debate. Welfare states have historically justified support on the basis of temporary misfortune, disability or old age, not permanent structural redundancy for large swathes of the population. Expanding unconditional transfers or universal basic income raises deep questions about deservingness, free-riding and social cohesion. Meanwhile, the prospect that young children today might never need to seek traditional employment collides with educational structures still geared towards preparing people for jobs that may not exist in twenty years.^6,10,15

Managing this transition requires more than economic engineering. It involves rethinking education towards lifelong learning, civic participation and creative skills; reshaping urban design and community institutions to accommodate more unstructured time; and constructing new narratives of status and contribution that are not anchored solely in paid employment. The risk is a bifurcation between a small elite of AI owners and shapers and a much larger population living on transfers but with limited agency over the systems that govern their lives.

Debates, objections and empirical uncertainty

Not all economists and labour scholars accept aggressive timelines for AI-driven underemployment. Historical experience with automation shows that while specific occupations disappear, new roles emerge, and aggregate employment can remain robust or even expand.^5,14,20 The World Economic Forum, for example, has projected large net job creation over the next decade when accounting for new roles in AI, green technologies and care, even as millions of existing jobs are displaced.⁵ International bodies such as the OECD highlight both opportunities and risks, stressing the role of policy in shaping outcomes.¹⁴

Critics of the most ambitious automation forecasts raise several objections:

• Task complexity and tacit knowledge: Many jobs involve tacit, context-specific skills, emotional labour and physical presence that are harder to automate than pure information processing. Even if AI can handle 80\% of cognitive tasks, the remaining 20\% may still require humans on-site, limiting the degree of headcount reduction.^14,20
• Adoption frictions: Regulatory barriers, liability concerns, cultural resistance and integration costs can significantly slow deployment, especially in highly regulated sectors like healthcare, aviation and law.^8,14
• New demand channels: Lower costs can stimulate new demand, creating jobs in adjacent areas. Historical examples include the way automation in textile manufacturing eventually led to a much larger fashion and retail ecosystem.^5,14
• Policy dampening: Governments could choose to slow automation in critical sectors, use subsidies to encourage worker retention or mandate human involvement in key decisions.^8,11

Proponents of rapid-disruption scenarios respond that this time may be different because the technology goes after the cognitive core of professional work, is general-purpose across sectors, and scales with data and compute rather than bespoke physical investments.^1,10,13 They also point to the increasingly software-native nature of the economy, where new products and services are digital from the outset and thus trivially automatable once models are capable. The actual path is likely to reflect elements of both views, with sectoral heterogeneity: some industries may see explosive AI-driven restructuring; others may evolve more slowly under the weight of regulation and human preference.

Why coherent policy frameworks cannot wait

Across this debate, one point of convergence is the need for structured preparation. A number of AI labs, investors and policy institutions now call for pre-emptive economic frameworks that can be activated as conditions change: enhanced labour-market statistics to monitor AI displacement in real time; scalable unemployment insurance systems; pre-authorised fiscal measures that can automatically expand support when indicators breach thresholds; and standing plans for introducing more ambitious tools such as basic income or sovereign AI funds if unemployment passes specified levels.^1,8,11,17

This contingency planning approach treats AI labour disruption similarly to other systemic risks: you do not wait for the flood to finish before designing the levees. Proposals include multi-scenario playbooks where, for instance, a 5\% unemployment scenario triggers one package of training grants and wage insurance, a 10\% scenario triggers expanded income support and sectoral transition programmes, and an unprecedented underemployment scenario opens the door to new forms of income replacement and capital redistribution.^8,11 The emphasis is on building institutional muscles now, while labour markets remain mostly intact, rather than scrambling under crisis conditions.

In parallel, educational, corporate and community actors have roles to play. Workforce frameworks focused on AI readiness urge universal AI literacy, worker participation in technology deployment and flexible training pathways that can be updated as task requirements shift.^2,5,20 Firms are encouraged to design AI adoption strategies that improve job quality and safety where possible, redeploy workers rather than simply shedding them, and share productivity gains in ways that maintain social licence.^2,8 And civic discourse needs to move beyond binary narratives of utopia or dystopia towards practical questions of ownership, governance and distribution.

The underlying claim that a large portion of economically valuable work may soon be performed by machines is ultimately less a prophecy than a forcing function. It surfaces uncomfortable but unavoidable questions about how societies tax, spend, educate and define human flourishing in an age of rapidly advancing intelligence technologies. Whether or not the most aggressive timelines prove accurate, the downside risk of being unprepared for a sharp labour-market shock is large, while the upside of having a coherent framework ready is substantial. In that sense, the real wager is not about the exact percentage of jobs affected, but about whether institutions can learn to move at something closer to the speed of code.

 

References

1. “We will need a new tax code for the wealth AI creates” – https://www.ft.com/content/b277360e-bf23-4366-afd7-acab940f66b7

2. Vinod Khosla Says AI Could Do ‘80% Of All Jobs’ By 2030 … – Finviz – 2026-03-05 – https://finviz.com/news/330419/vinod-khosla-says-ai-could-do-80-of-all-jobs-by-2030-reshaping-15-trillion-us-labor-economy

3. The AI-Ready Workforce – Jobs for the Future (JFF) – 2023-10-05 – https://info.jff.org/ai-ready

4. Vinod Khosla Says AI Could Do ‘80% Of All Jobs’ By 2030 … – 2026-03-11 – https://finance.yahoo.com/news/vinod-khosla-says-ai-could-094605137.html

5. VC Vinod Khosla predicts AI will replace 80% of jobs by 2030 – 2025-07-02 – https://www.linkedin.com/posts/analytics-india-magazine_venture-capitalist-vinod-khosla-believes-activity-7346129524771758080-l66t

6. AI and the Future of Work: A Strategic Framework for Skills … – 2025-05-08 – https://www.linkedin.com/pulse/ai-future-work-strategic-framework-skills-development-zaman-ljhuc

7. Vinod Khosla Predicts AI Will Eliminate 80% of Jobs – LinkedIn – 2026-03-04 – https://www.linkedin.com/posts/fortune_openai-investor-vinod-khosla-predicts-today-activity-7435021301309837312-SeGz

8. Silicon Valley investor Vinod Khosla predicts AI will replace 80% of … – 2025-07-01 – https://finance.yahoo.com/news/silicon-valley-investor-vinod-khosla-205800150.html

9. [PDF] Industrial Policy for the Intelligence Age: Ideas to Keep People First – https://cdn.openai.com/pdf/561e7512-253e-424b-9734-ef4098440601/Industrial%20Policy%20for%20the%20Intelligence%20Age.pdf

10. Vinod Khosla on the End of Jobs and the Future of Capitalism – 2026-05-12 – https://podcast.newcomer.co/episode/vinod-khosla-on-the-end-of-jobs-and-the-future-of-capitalism

11. OpenAI investor Vinod Khosla believes AI will be able to do 80% of … – 2026-03-06 – https://fortune.com/2026/03/06/vinod-khosla-predicts-80-percent-of-jobs-done-by-ai-15-trillion-of-gdp-going-away/

12. Policy on the AI Exponential – Anthropic – 2026-06-10 – https://www.anthropic.com/policy-on-the-ai-exponential/epf

13. Vinod Khosla Predicting the Future | Ep. 15 – YouTube – 2025-07-01 – https://www.youtube.com/watch?v=KZ9cYDeum4U

14. Khosla warns of disrupted decade before AI boom brings abundance – 2025-09-18 – https://www.top1000funds.com/events/fis/fis-stanford-2025/khosla-warns-of-disrupted-decade-before-ai-boom-brings-abundance/

15. Future of work – OECD – 2019-04-25 – https://www.oecd.org/en/topics/future-of-work.html

16. The End of Work: Vinod Khosla’s Bold AI Prediction – YouTube – 2026-03-04 – https://www.youtube.com/watch?v=cSWvm7nu1rI

17. OpenAI investor Vinod Khosla predicts AI could wipe out jobs but … – 2026-03-04 – https://fortune.com/videos/watch/OpenAI-investor-Vinod-Khosla-predicts-AI-could-wipe-out-jobs-but-make-life-better-by-2040/f083e466-3571-4596-8524-7fb2a4ba1aea

18. AI and the Future of Work: A Policy Framework for Transforming Job … – 2021-01-16 – https://www.academia.edu/44919138/AI_and_the_Future_of_Work_A_Policy_Framework_for_Transforming_Job_Disruption_into_Social_Good_for_All

19. Vinod Khosla forecasts AI job displacement but improved living … – 2026-04-06 – https://www.instagram.com/reel/DWyhLLBAg4o/

20. Vinod Khosla’s Warning for India’s IT Industry | Can AI Save It? – 2026-06-13 – https://www.youtube.com/watch?v=pfGAdzo09xw

21. AI futures of work – Cedefop – European Union – 2018-03-13 – https://www.cedefop.europa.eu/en/projects/digitalisation-and-future-work

22. Vinod Khosla says most modern work is a form of servitude. AI will … – 2025-07-03 – https://www.reddit.com/r/singularity/comments/1lqn9h6/vinod_khosla_says_most_modern_work_is_a_form_of/

23. Vinod Khosla says AI will take over 80% of work in 80% of jobs – 2024-09-25 – https://www.reddit.com/r/OpenAI/comments/1fos72b/vinod_khosla_says_ai_will_take_over_80_of_work_in/

24. “BPO and IT industries will vanish” – Vinod Khosla makes bold AI … – 2010-01-01 – https://www.facebook.com/EconomicTimes/posts/-bpo-and-it-industries-will-vanish-vinod-khosla-makes-bold-ai-prediction-%EF%B8%8F-on-ni/1214443320711574/

25. Societal Innovation – Khosla Ventures – https://khoslaventures.com/entrepreneurs/societal-innovation

26. Famed Silicon Valley investor Vinod Khosla says universal basic … – 2024-09-24 – https://www.reddit.com/r/Futurology/comments/1fo9p0i/famed_silicon_valley_investor_vinod_khosla_says/