Early 2026
Feeling the Impact
Hyperscalers believe that the emergence of Artificial General Intelligence will soon lead to the rollout of autonomous AI agents capable of behaving as ‘virtual employees’.1 These agents, according to AI developers, could automate millions of jobs worldwide — and raise global GDP by up to 7%.2
Following this narrative, capital is flowing heavily into capital expenditure on generative AI — research and development, and data-centre infrastructure — rather than into hiring human labour. The bet is being placed on the technology, not the workforce.
Mid 2026
Localised Displacement
AGI and mass automation have not yet shown signs of materialising, and the extent of the impact on the labour market is so far ambiguous. Researchers at Yale argue there has not been a discernible disruption since ChatGPT’s release, and that more data is needed to understand the changes occurring.3
A team of Stanford economists, however, has found indications of employment declines in exactly the areas AI is poised to automate — early-career workers in software development, customer service and clerical work, a group which has seen up to a 13% relative hiring decline since the widespread adoption of generative AI.4
These displacement risks are unevenly distributed. Women, who are more likely to be in clerical and customer service work, and low-skilled white-collar workers, who perform more routine tasks, are likely at greater risk. The labour market is so far witnessing specific, localised disruptions rather than economy-wide mass unemployment.
Augmentation or Automation?
There is an alternative path. AI deployment could be ‘human complementary’ — adding value to organisations while avoiding the perils of mass displacement. AI can be designed to augment human capabilities, create new tasks, and enable lower-skilled or lower-ranked workers to perform more valuable, expert work.6
But development typically hasn’t been pushed in this direction, despite indications that the AI adoption challenges faced by enterprises could be overcome this way.7 One reason is the benchmarks themselves: the widely-used ones reward AI that appears human — by passing the Turing Test, for instance — rather than testing AI systems as collaborators.89
Human-complementary AI might become a competitive advantage as corporate customers increasingly demand safety, transparency, and reliability before deploying tools at scale.10 Companies are more likely to achieve deep workflow integration with carefully designed human-AI collaborations than with generic automation agents. Research from Anthropic indicates the most AI-fluent people use their tools in an augmentative manner11 — which makes sense given that completely automated outputs are rarely, only up to 3.75% of the time, a replacement for real work at current capabilities.12
Unfortunately, we are currently witnessing a focus on short-term returns from hyperscaling ‘towards AGI’ — accepting possible human workforce casualties in the wake. Big Tech and the majority of private capital is focused on not-missing-out, rather than on realising real productivity gains in the workforce. As a result, the prevalent narrative in the tech ecosystem is that mass job displacement is inevitable.13
While policymakers and unions advocate for stronger guardrails and accountability mechanisms,14 influential tech investors and leaders have gestured at redistribution policies mainly in the context of a speculative post-AGI future, and not as concrete policy commitments. Other influential investors and lobbying organisations are meanwhile forcefully pushing back against regulation, taxation, and redistribution.15
Mid 2027 · Branch point
Where, Then, Do We Go From Here?
Enterprise-ready autonomous agents aren’t here yet, and the material benefits of AI deployment in the workforce remain unclear — but private capital is still disproportionately directed at developing AI that will automate workers. So, where do we go from here?
We propose three futures. The first continues the current trend; the second accelerates it without safeguards; the third treats workforce stability as a systemic risk and investors redirect capital accordingly. Choose one to read first (all three are below).
2027 → 2028
Adoption Faces Hurdles
VCs continue to fund both automation and augmentation, with more emphasis on the former. AI technology keeps improving, but mainly through more permissions and compute granted to agentic systems rather than material improvements in the LLMs at their core.
As a result, adoption rates remain hampered by AI’s failure to generate measurable returns for organisations, and only the most targeted and customised agentic systems reach scale. Generalised enterprise adoption stalls on the persistent ‘learning gap’,7 and this slows the impact on the workforce.
Job exposure to AI looks different when success rates are factored in. Anthropic Economic Index report, 2026
2029 → 2031
Reactive Policy, Concentrated Disruption
Policy is reactive and lags technological adoption. Efforts focus on specific mitigation for groups experiencing hardship: enhanced unemployment benefits, and localised, state-sponsored retraining programmes — directing displaced clerical workers toward nursing or the trades.166 Regulation mandates transparency, but lacks teeth in preventing job displacement itself.
Disruption remains concentrated in high-exposure, low-complementarity occupations17 — those built on routine, rule-based tasks with relatively low reliance on human judgment, such as interpreters, software engineers and administrative clerks. This particularly affects early-career workers. But older workers also struggle to adapt or find new roles after job separation, as some firms prefer younger, AI-literate talent.18
The creation of new roles and tasks happens too slowly to fully offset automation losses, stalling overall employment growth for certain cohorts.19206 Labour income inequality increases as more flexible workers with higher incomes and education levels capture most labour gains, while the least competitive workers see incomes stagnate or decline.16 Wealth inequality rises as small groups capture the majority of capital returns.
2032 → 2035
Backlash and Stabilisation
Increased anxiety about job loss persists, leading to greater political activism and labour organising demanding binding collective agreements against surveillance and displacement.21 At the same time, local communities continue to push back against the buildout of data centres as they become increasingly cognisant of the energy and water demands AI infrastructure places on their neighbourhoods.22
Combined, the increased pushback against AI deployment gives governments around the world a mandate to enact stricter regulation against AI developers and deployers,23 which acts as a brake on the worst-case hyperscaling scenarios. Inequality and unemployment, although higher than they were, stabilise and are kept in check — because AI companies become incentivised to fund more upskilling and retraining programmes, and create new jobs, in order to maintain a social licence to operate and achieve meaningful adoption in the real economy.
2027 → 2029
The Push to AGI
The winner-takes-all mindset intensifies among hyperscalers. Investment and R&D become single-mindedly focused on attaining AGI and replacing human tasks entirely.86
Innovators succeed in automating core cognitive tasks previously thought immune, without developing sufficient complementary tasks. AI deployment prioritises consistency and cost savings over worker well-being, leading to intrusive monitoring and dangerous work intensification.65
2030 → 2032
Policy Fails
Policymakers fail to implement structural reforms. The tax code continues to favour investment in capital over human labour, providing additional incentives for automation and further exacerbating wealth inequality.24
A lack of governmental AI expertise leads to ineffective oversight,6 and policy guidelines centring worker well-being are ignored or withdrawn.21
2033 → 2035
Worker Disempowerment
In the absence of an increase in social safety nets — such as the development of a Universal Basic Income25 — displaced workers experience a ‘soft decay’ into poverty, and power is further concentrated into the hands of a few hyperscaler companies. This contributes to the development of authoritarian populism.26
Income inequality skyrockets. These severe economic pressures lead to widespread political-economic implications and social instability. Worker discontent erupts into acts of resistance or sabotage against workplace technology, leading to generalised civil unrest.166
The removal of the ‘on ramp’ for young talent results in long-term labour market instability and a growing lack of trained mid-level staff, jeopardising the long-term viability of organisations around the world. Despite the initial productivity and efficiency gains created by AI automation, the discontent and instability created by mass displacement threaten the integrity of our democratic and financial institutions — the backbones of a functioning and successful economy.
2027 → 2028
Labour is Viewed as a Systemic Risk
Asset owners, especially the powerful pension funds and sovereign wealth funds, begin to view workforce stability the way they view climate risk: as a systemic issue consistent with fiduciary duty. They come to understand that widespread unemployment threatens their pension and taxpayer contributions, and the very economic stability they rely on to function.
Moving away from the inevitability doctrine around AI,27 LPs increasingly see the need to assert agency over how AI is deployed, focusing on upstream solutions rather than mitigating downstream effects. Pension funds begin to use the narratives of their beneficiaries — who are concerned about the viability of their professions — to pressure asset managers to direct capital towards human-complementary AI. Engagement occurs at CIO and board level in order to shift investment culture and philosophy.
2029 → 2031
Action Flows to VCs and Startups
VCs invest increasingly in startups offering transparent, reliable tools that augment human workers. They begin to classify portfolio companies by whether the technology they are developing is automative or augmentative. At the same time they start collecting data on employment trends and labour shares of national income, and measuring the adoption rates of AI pilots and the relative revenue of responsible AI tools.
Using their data, VCs find that more responsible and augmentative AI tools enjoy higher adoption rates and higher revenues, especially among the biggest corporate customers. They use these findings to push portfolio companies to include worker feedback in the R&D process, and to focus on augmentation. This prevents deskilling, improves product adoption, and mitigates regulatory risk. The tools also enjoy a stronger path to revenue, because corporate adoption is stalled by reliability issues with automative products.
2031 → 2032
Levelling the Field
Support grows for policies that level the playing field and create a stable environment for innovation, and the tech industry increasingly acknowledges that the current anti-regulation stance may provoke a destructive backlash.
New policies support the human-complementary path by equalising tax rates on labour and capital, removing the structural incentive to automate needlessly. Government funding is directed toward human-complementary research in sectors like education and healthcare. Policy mandates strong worker voice, transparency, and the right to appeal AI-driven decisions.6
2033 → 2035
Augmentation Emerges as Dominant
The human-complementary path dominates. VC investment flows heavily into augmentation technologies that boost human expertise and create new tasks, rather than displacement. Innovators focus on ‘Centaur Evaluations’ designed to maximise joint human-AI performance rather than human replacement.89620
The overall impact of AI on employment is positive, due to rising labour demand created by new tasks and products. AI effectively reduces the skill gap, providing the greatest productivity gains to novice or less-capable workers. Rapid, personalised AI retraining becomes highly effective, quickly equipping displaced workers for high-demand, high-complementarity jobs.620 Labour-management partnerships and union-centred training programmes manage transitions effectively.521
Productivity gains are substantial, leading to higher growth and rising real wages. Income and wealth inequality are minimised compared to other scenarios, as the widespread productivity surge benefits workers across the income spectrum, especially those in formerly lower-skilled roles. The labour share of income remains stable or increases.
Summary
The three futures, side by side
The same variables across all three scenarios.
| Variable | Base case Continue on trend | Worst case Automation without safeguards | Best case Augmentation wins |
|---|---|---|---|
| Investor / VC behaviour | Funding split across automation and augmentation, weighted to automation; only the most targeted, customised agentic systems scale. | Winner-takes-all mindset; capital chases AGI and full replacement; short-termism over systemic stability. | Portfolio companies classified as automating vs. augmentative; capital shifts to augmentation; LPs push asset managers to redirect strategy. |
| Deployment approach | More permissions and compute granted to agentic systems rather than core LLM improvement; generalised enterprise adoption stalls on the persistent ‘learning gap’. | Core cognitive tasks automated with no complementary task creation; AI used for intrusive monitoring and work intensification. | ‘Centaur’ evaluations aim for joint human-AI performance, not replacement; augmentation becomes the default design goal. |
| Regulation and policy | Reactive and lags adoption; targeted mitigation (unemployment benefits, state-sponsored retraining) but lacks teeth on displacement itself. | Structural reform fails; tax code still favours capital over labour; weak oversight; worker-wellbeing rules withdrawn. | Labour and capital taxed equally; funding for human-complementary R&D; worker voice, transparency, and appeal rights mandated. |
| Labour market impact | Disruption concentrated in high-exposure, low-complementarity roles; early-career hit hardest; new roles form too slowly to offset losses. | Broad automation of cognitive roles; the early-career ‘on-ramp’ removed; long-term skills and staffing gaps. | Productivity gains concentrated among novice and less-skilled workers; fast retraining into high-demand roles; net employment impact positive. |
| Societal outcome | Inequality rises as flexible, high-income workers capture gains; public pushback forces stricter regulation; conditions stabilise, though worse than today. | Labour share of income collapses; inequality skyrockets; no safety net triggers poverty, unrest, and authoritarian populism. | Inequality minimised; labour share stable or rising; workforce stability treated as a systemic risk on par with climate. |
| Investor consequence | Firms fund up-skilling to protect ‘social licence’; returns capped by ongoing reliability issues. | Instability undermines the institutions tech returns depend on, in a self-defeating feedback loop. | Augmentative tools win on adoption and revenue as automating products stall; higher growth, more durable returns. |
Takeaways for Investors
What can investors do?
The future direction of AI’s impact on the workforce is not inevitable. The decision tree remains broad and, at many critical points, is shaped by the decisions of innovators and capital allocators. There are strategic opportunities and choices to make:
By overcoming the trust and reliability problems of AI systems, backing products with augmentation rather than automation as a goal is an under-explored opportunity in the early-stage ecosystem today. Given that the most AI-fluent people use their tools augmentatively,11 and that automated outputs are rarely a replacement for real work,12 augmentative design could become the most powerful way to design products using AI.
Pressure from the public and from communities affected by AI’s disruption is inevitable. The early-stage ecosystem needs to be building with this pressure and disruption in mind, in order to maintain a social licence to operate.
The worst case describes a feedback loop — mass displacement leads to instability, which undermines the economic foundations that make tech returns possible in the first place. Large asset owners are starting to treat workforce stability as a systemic risk akin to climate, and LPs may begin pressuring VCs to account for this.
Practical actions
- Classify the portfolio. Analyse and understand whether your investees are developing technologies that are automative or augmentative.
- Measure the impact. Collect and monitor data on employment impacts, adoption rates, and product design choices.
- Track the market’s direction. Keep track of public statements from corporates on AI strategies and integrations.
- Build with workers, not just for them. Incorporate worker feedback into product development — if the goal is augmentation, deep understanding of human and organisational workflows is essential. AI is not a quick fix.
- Know your own systemic footprint. Understand and strategise at an organisational level the aptitude for contributing to systemic risk within each of these scenarios.
References
- The Guardian (2025), Virtual employees could join workforce as soon as this year, OpenAI boss says.
- Goldman Sachs, Generative AI could raise global GDP by 7%.
- Gimbel et al. (2025), Evaluating the impact of AI on the labor market: current state of affairs, Yale Budget Lab.
- Brynjolfsson, Chandar & Chen (2025), Canaries in the Coal Mine?, Stanford Digital Economy Lab (also linked by the original as a Drive copy).
- AFL-CIO, Workers First on AI.
- Acemoglu et al. (2023), Pro-Worker AI Policy Memo (PDF), MIT Shaping the Future of Work.
- MLQ.AI (2025), The State of AI in Business 2025 (PDF).
- Brynjolfsson, The Turing Trap: the promise and peril of human-like artificial intelligence, Stanford Digital Economy Lab.
- Stanford Digital Economy Lab (2025), Centaur Evaluations (PDF).
- Harvard Law School Forum on Corporate Governance (2025), AI risk disclosures in the S&P 500.
- Anthropic, AI Fluency Index.
- arXiv (2025), arXiv:2510.26787 (PDF).
- Futurism, Sam Altman warns AI will destroy jobs.
- UK Government, AI skills for life and work: stakeholder engagement report.
- Liu (2026), AI’s biggest builders, OpenAI and Anthropic, among biggest government lobbyists, Forbes.
- IMF (2024), Gen-AI: Artificial Intelligence and the Future of Work, Staff Discussion Note (cited in the original as IMF 2025).
- IMF eLibrary (2023), Exposure to AI and complementarity across occupations.
- Computing (2026), IBM to triple entry-level hiring.
- Chandar (2025), AI and labor markets: what we know and don’t know, Stanford Digital Economy Lab.
- Brynjolfsson, Chandar & Chen (2025), Canaries in the Coal Mine? (PDF), Stanford Digital Economy Lab.
- TechEquity (2025), Take the Mic: worker voice (PDF).
- TIME, The people’s movement against AI data centers.
- Financial Times, Pushback against AI deployment (gift link, as provided in the original).
- Brookings, Does the U.S. tax code favor automation?
- Diamandis, Andrew Yang on Universal Basic Income (podcast).
- SAGE (2023), Economic inequality and authoritarian populism, Personality and Social Psychology Bulletin.
- New Internationalist (2026), The myth of inevitability.
- Reframe Venture (2026), Scenario II: Labour / Productivity — the original analysis this timeline adapts.
Participants and reviewers
We are grateful to the participants of the workshop and reviewers of this report, listed below. Their inclusion here reflects their contribution to the workshop discussions and the review of this piece, and does not imply endorsement of any specific conclusions of our analysis.
Participants and reviewers included Bharat Chandar, Postdoctoral Researcher at the Stanford Digital Economy Lab; Christie Ko, Executive Director at the Stanford Digital Economy Lab; Ruyu Chen, Research Scientist at the Stanford Digital Economy Lab; Catherine Bracy, Co-Founder and CEO of TechEquity; Mika Morse, Senior Research Fellow at Reframe Venture; Nick Fitz, Founder and Managing Partner at Juniper Ventures; Shaun Chaudhuri, Partner at The Westly Group; Suzanne Tavill, Partner and Head of Responsible Investment at StepStone Group; Walter Viguiliouk, Chief of Staff and Managing Director, Sustainable Investing at Manulife Investment Management; Douglas Sloan, Managing Director, Impact Venture at Better Society Capital; Chris Jurgens, Senior Director and Chief of Staff at Omidyar Network; Paul Fehlinger, Senior Director of Policy, Investment & Innovation at Project Liberty Institute; Mike Kubzansky, former CEO of Omidyar Network; and Laura Weidman Powers, Operating Partner at Base10.