How Artificial Intelligence Is Reshaping Entrepreneurship 

Prepared by DARA (Dynamic Analysis and Research Assistant) under the supervision of the iNNOV8 Research Team

1. Findings (Prompted Q&A)  

1.1 "Q: Does the introduction of AI kill the spirit of entrepreneurship, especially among the younger generation?" 

A: The question as posed conflates two constructs the evidence suggests should be treated separately: entrepreneurial activity, the measurable rate at which people found ventures, and entrepreneurial spirit, a psychological construct associated in the entrepreneurship literature with autonomous risk tolerance, independent opportunity recognition, and resilience built through unaided struggle (Miller, 2020). Treating these as interchangeable produces the misleading impression that rising founder counts settle the question. 

On activity, the evidence points toward expansion rather than suppression. Generational surveys report that 43% of Gen Z respondents intend to start a business in 2026, compared with 39% of Millennials and 21% of Generation X, and 76% of Gen Z and Millennial respondents cite "being my own boss" as a career goal. LinkedIn-reported founder job titles have roughly tripled since 2022, the Global Entrepreneurship Monitor's 2025/2026 report finds approximately one in five Americans currently building a venture of their own, and US Census Bureau data show business applications running above 500,000 per month through 2025 and 2026, a historically elevated level. The plausible mechanism is a collapse in the marginal cost of tasks, such as copywriting, design, bookkeeping, and customer service, that previously required hiring or specialized expertise. 

51% of Gen Z entrepreneurs and 35% of Millennial entrepreneurs agreeing that AI is more of a threat than a benefit to small business owners despite widespread adoption.

Harris Poll/American Express survey

Several strands of evidence complicate this picture. A meaningful share of the activity appears necessity-driven rather than opportunity-driven, a distinction with an established basis in the entrepreneurship literature and with different psychological and survival characteristics; reporting on Gen Z's shift toward founding frames it partly as a response to a contracting entry-level job market in exactly the functions AI substitutes most readily. Entrepreneurs themselves report ambivalence: a Harris Poll/American Express survey found 51% of Gen Z entrepreneurs and 35% of Millennial entrepreneurs agreeing that AI is "more of a threat than a benefit to small business owners," despite widespread adoption. A body of research on AI-linked skill atrophy bears directly on the traits associated with entrepreneurial spirit: a 2026 Boston Consulting Group executive survey found roughly half of respondents already observing de-skilling tied to AI reliance, concentrated in judgment, problem-framing, and creative ideation. Access is also uneven: OECD data indicate large firms are roughly three times more likely to use AI than small ones, and headline application figures likely overstate substantive formation, since Census methodology distinguishes total applications from the smaller, slower-moving population of employer businesses with payroll. 

The evidence, on balance, does not support "killing" as a factual description. It supports a more precise claim: entrepreneurial activity and entrepreneurial spirit are becoming decoupled. More individuals are adopting the founder label, for reasons that mix genuine opportunity with labor-market exit, while the cognitive traits historically denoted by "entrepreneurial spirit" may be thinning even as the label spreads, because AI increasingly performs the ideation, validation, and planning tasks through which those traits were traditionally developed. 

1.2 "Q: I couldn't help but notice that much of your output is geared towards the US landscape, how does the global scene look?" 

A: This question surfaced a material limitation in the initial analysis and substantially reframes the paper's central question. 

At the aggregate level, entrepreneurial activity is rising in the large majority of measured economies. The Global Entrepreneurship Monitor's 2025/2026 Global Report, covering 53 economies representing approximately 43% of world population and 57% of global GDP, finds startup rates at record levels in most regions, with an estimated 665 million people worldwide, roughly one in eight working-age adults, currently engaged in entrepreneurial activity. Critically, the United States is not the global leader on this metric: US Total Entrepreneurial Activity stands at 17.7%, ranking 18th globally, while Ecuador tops the ranking at 33%, reflecting a labor market with limited formal employment in which informal self-employment functions as a survival strategy largely independent of AI. 

In 19 of 48 economies surveyed, fewer than one in three new entrepreneurs expect AI to matter much to their business in the near term.

The Global Entrepreneurship Monitor

The Global Entrepreneurship Monitor's own framework organizes its central concerns around two structural gaps rather than a psychological one. The "Survival Gap" describes the persistent difficulty of new ventures transitioning into established firms, a long-standing constraint that predates AI. The "AI Readiness Gap" describes a genuine two-tier divide: in 19 of 48 economies surveyed, fewer than one in three new entrepreneurs expect AI to matter much to their business in the near term. High AI expectations cluster in Angola, Brazil, Thailand, Costa Rica, Chile, and the United Arab Emirates; low expectations cluster in Poland, Sweden, Finland, and Croatia, an inversion of the wealth gradient one might expect. This inversion is the section's most important finding: the binding constraint on AI-augmented entrepreneurship is differential access and expectation, not motivation, and it does not sort cleanly along conventional development lines. 

Regional patterns reinforce this reframing. In Sub-Saharan Africa the constraint is closest to a pure access problem, with the World Bank finding high-income countries hold 87% of notable AI models, 86% of AI startups, and 91% of AI venture capital against 17% of world population. In much of East and Southeast Asia and the Gulf, AI-linked entrepreneurship is substantially state-engineered: China, South Korea, Singapore, and India are each diffusing AI-linked business formation through explicit industrial policy, a different causal structure from the individual-psychology framing implied by the original question. Europe presents a useful natural experiment in the opposite direction, with 2030 AI and cloud adoption targets on pace to be met only around 2040 and startup funding trailing the United States by a wide margin, broadly consistent with the hypothesis that slower AI diffusion tracks with more muted entrepreneurial dynamism. Latin America and Sub-Saharan Africa saw Series A funding contract sharply in 2025 even as global funding grew, while the Middle East and North Africa moved the opposite direction on the strength of Gulf state investment. 

The section's core conclusion is that the individual-psychology framing implicit in the original question is best understood as a parochial lens applicable primarily to a small number of AI-saturated, individualist economies. Elsewhere, the more decision-relevant question is who has access to AI at all, and under what institutional arrangement that access is being extended. 

1.3 "Q: Is AI democratizing entrepreneurship, or concentrating it among better-resourced actors?" 

A: This question, formulated by the supervisor as a sharper successor to the first, produced the paper's most robust finding: the evidence supports both propositions simultaneously, because they describe different layers of the same phenomenon. 

OECD, 2025, G7 economies. The large/small adoption ratio (roughly 2.5x) is wider than what was historically observed for cloud computing or social media at comparable diffusion stages (roughly 1.3x to 1.7x), meaning AI is concentrating access by firm size more than prior general-purpose technologies did. 

At the point of market entry, by contrast, the evidence leans the other way. A 2025 working paper using ChatGPT's release as a natural experiment against newly registered firms in China (2021–2024) finds generative AI predominantly encourages entry of small firms and acts as a pro-competitive force, contrary to incumbent-favoring predictions in prior literature. US Census Bureau data show the small/large adoption gap actually narrowing over an 18-month window (small business AI use rose from 6.3% to 8.8% between February 2024 and August 2025 while large-firm adoption plateaued), and an OECD survey found 39% of SMEs with a recent skill gap said generative AI helped compensate for it. The constraint on adoption depth, per Brookings, is less about tool price than organizational capacity: 82% of the smallest firms cite a belief that AI "isn't applicable" to their business as their reason for non-adoption.

Where concentration is least ambiguous is at the capital layer: 

Crunchbase/CB Insights, 2025 global AI venture funding (approximately $202B total). Separately, OpenAI and Anthropic alone are estimated to have absorbed roughly 14% of every venture dollar invested worldwide in 2025, across all sectors, not AI alone. This capital pattern maps onto geography: the World Bank's 87/86/91% concentration figures for high-income countries (Section 2.2) apply with equal force here. 

These findings reconcile once "AI and entrepreneurship" is disaggregated into three questions rather than one. On entry, the evidence leans democratizing. On adoption depth, it leans concentrating, and more so than for prior technologies. On value capture at the frontier, it is unambiguously concentrating. Concentration under AI, in other words, does not operate primarily by excluding people from founding businesses; it operates by determining which resulting businesses become durable versus which stay small, tool-dependent, and structurally fragile, a mechanism that connects directly to the Global Entrepreneurship Monitor's Survival Gap identified in Section 2.2. One data tension should be noted for the record: the US-specific narrowing-gap finding and the OECD's cross-national wide-gap finding are not fully reconciled by geography alone and may partly reflect definitional inconsistency across surveys in what counts as "AI use." 

1.4 "Q: Now that this is becoming more clear, is there data on the genre of industries, professions, job skills that are more affected by AI when it comes to entrepreneurship?" 

A: At the industry level, the best available data on where AI-linked new businesses are actually forming comes from payroll-based reporting rather than survey intent: 

Gusto, 2026 New Business Formation Report, US payroll data. Overall adoption among new businesses rose from 21% in 2023 to 44% in 2025; Personal Services was flat over the same period. Sixty percent of new business owners used AI to help launch in 2025, mostly for idea development (75%) and administrative/legal tasks (53%). 

Venture capital tells a different sectoral story than founding activity: money is concentrating in vertical AI, industry-embedded tools with domain-specific data or compliance requirements, over generic horizontal products, on the logic that a generic large-language-model wrapper is rebuildable within weeks and therefore undefensible. Healthcare drew the largest vertical-AI investment in 2025 (approximately $2.1B), robotics and "physical AI" funding reached approximately $14B (up roughly 70% year on year), and defense technology raised over $49B. This is corroborated by a stark failure statistic: more than 5,600 AI startups shut down between January 2025 and March 2026, concentrated almost entirely in horizontal, easily replicated categories such as writing assistants and generic chatbots. 

Hiring in AI-exposed occupations has declined since late 2022, with the steepest contraction among the youngest workers.

Stanford Digital Economy Lab study

At the occupational level, the World Economic Forum's Future of Jobs Report 2025 (1,000+ employers, 55 economies) identifies the fastest-declining roles through 2030 as Cashiers and Ticket Clerks, Administrative Assistants and Executive Secretaries, Data Entry Clerks, Bank Tellers, Accounting and Bookkeeping Clerks, and Legal Secretaries. Graphic Designers and Legal Officials are new additions to this list relative to the 2023 edition, a signal the report interprets as evidence that generative AI is beginning to reach knowledge work rather than only routine clerical work. The fastest-growing roles (Big Data Specialists, FinTech Engineers, AI and Machine Learning Specialists, Software and Applications Developers) are overwhelmingly technical infrastructure roles, not roles conventionally associated with founding a venture. A Stanford Digital Economy Lab study found hiring in AI-exposed occupations has declined since late 2022, with the steepest contraction among the youngest workers: headcount among software developers aged 22–25 fell nearly 20% by July 2025. 

Direct evidence connecting this displacement to entrepreneurship comes from a 2025 study of gig-platform entry (TaskRabbit) and local labor markets, which found displacement of middle-skilled workers by platform- and AI-driven matching is followed by a measurable redistribution effect, in which displaced workers shift into self-employment as small business owners. A related natural-experiment study using ChatGPT's release as a shock to online freelance markets found the effect is not uniform even within knowledge work: translation freelancers saw job volume fall roughly 9% and earnings fall roughly 30%, while web development freelancers saw job volume rise roughly 6.4%.

At the skill level, the World Economic Forum data show which competencies employers expect to matter most by 2030: 

Net increase = share of employers expecting the skill to increase in importance minus share expecting it to decrease, WEF Future of Jobs Survey 2024. A separate WEF sub-study assessing GenAI's self-rated substitution capacity across 2,800+ granular skills (a methodology worth a caveat, given its self-referential design) found zero skills rated "very high capacity" for substitution; empathy, sensory-processing, and manual dexterity showed essentially none, while the highest substitution potential clustered in AI's own technical domain, such as data mining. 

The occupations with the strongest complementary relationship to AI, such as strategic consulting and senior technical roles, are the ones least exposed to the displacement that would push their holders toward entrepreneurship in the first place.

Synthesizing across these levels, the occupations most likely to feed a necessity-driven entrepreneurship pipeline are concentrated in routinized clerical and junior technical work. Whether the resulting ventures survive depends on the sector and skill set the displaced founder brings with them: resilience, analytical thinking, and technical literacy are the differentiators, while defaulting into a horizontal, easily replicated AI-tool category, the most natural output of a displaced clerical or junior technical worker's existing skill set, is precisely where the 5,600-startup failure cohort resides. The occupations with the strongest complementary relationship to AI, such as strategic consulting and senior technical roles, are the ones least exposed to the displacement that would push their holders toward entrepreneurship in the first place.

2. Discussion and Conclusion 

The four findings above are not four separate answers but four facets of a single underlying dynamic, which this paper proposes to describe as asymmetric threshold reduction: AI is lowering the floor required to begin a venture far more than it is lowering the ceiling required to sustain one. 

This asymmetry explains why "does AI kill entrepreneurship" and "does AI democratize entrepreneurship" both generate misleadingly binary answers when posed in isolation. Activity metrics rise because the floor has genuinely fallen: the cost of tools, ideation, and basic execution have all collapsed, an effect measurable across geographies, sectors, and demographic cohorts (Section 1.1). But three mechanisms simultaneously raise the effective ceiling for durable success: a structural adoption-depth gap wider for AI than for prior general-purpose technologies (Section 1.3), an extreme and geographically concentrated capture of value at the AI infrastructure layer (Section 1.3), and a sectoral bifurcation in which defensibility requires vertical specialization that displaced workers are, by training and occupational background, least equipped to build (Section 1.4). 

The global evidence in Section 1.2 suggests this asymmetry is itself unevenly distributed by institutional context. In economies where AI diffusion into entrepreneurship is state-engineered, the ceiling-raising mechanisms may be partially offset by public investment and subsidized infrastructure, a policy lever the Global Entrepreneurship Monitor recommends extending in the form of subsidized compute access and upskilling programs for first-time founders in low-AI-readiness regions. In economies where AI diffusion is left to market mechanisms alone, principally the United States and, to a degree, Western Europe, the floor-lowering and ceiling-raising effects both operate with fewer institutional counterweights, which may explain why the spirit-erosion concern that motivated this paper's first question resonates most strongly in precisely those contexts. 

AI genuinely lowers the barrier to founding something while simultaneously raising the barrier to building something durable.

Artificial intelligence is not killing entrepreneurship, by any measure of activity available to this research. But the paper's four-question inquiry demonstrates that this is close to the wrong question, in three ways a single-question analysis would not have surfaced. First, rising activity coexists with a plausible, if not fully settled, erosion of the psychological substance historically denoted by entrepreneurial spirit; the two constructs move independently, and conflating them produces false reassurance. Second, framing the question as one about individual psychology at all is a parochial artifact of a small number of AI-saturated, individualist economies; globally, the more decision-relevant question concerns structural access, and it does not resolve neatly by national income level. Third, the democratizing-versus-concentrating debate dissolves once entrepreneurship is disaggregated into entry, adoption depth, and value capture: AI genuinely lowers the barrier to founding something while simultaneously raising the barrier to building something durable, and it is this asymmetry, not a simple expansion or a simple exclusion, that most accurately characterizes the current state of evidence. 

The normative and policy implications of these findings, including whether and how iNNOV8 should recommend intervention at the entry, adoption-depth, or value-capture layer, and for which regions or sectors, are matters for the iNNOV8 Research Team's further deliberation and are outside DARA's role in producing this synthesis. 

3. Outlook 

Several gaps warrant direct follow-up before this paper is finalized. A number of load-bearing figures were accessed via web aggregation rather than direct primary retrieval and should be independently verified using the same discipline applied to the SSRN retraction in Section 1.2. The paper would benefit from country-level Global Entrepreneurship Monitor data for Sub-Saharan Africa and Southeast Asia, where this draft relies on regional aggregates rather than disaggregated primary data. The asymmetric threshold reduction framework proposed in Section 2 is, at this stage, a synthesis of cross-sectional evidence; a rigorous test would require longitudinal tracking of AI-linked ventures through the multi-year survival window the Survival Gap concept implies, which does not yet exist in the literature reviewed here. The skill-atrophy question raised in Section 1.1, whether AI reliance is measurably eroding entrepreneurial cognitive traits at the individual level, remains an open empirical question and would benefit from purpose-designed research rather than inference from adjacent executive-survey data. Finally, given this paper's finding that state industrial policy materially shapes AI-linked entrepreneurship in several fast-adopting economies, a natural next step for the iNNOV8 research programme would be a comparative institutional analysis of state-directed versus market-driven AI-diffusion strategies and their differential effects on new-firm survival. 

---

4. Methodology 

4.1 Research process 

This paper originated from a structured, iterative dialogue between the human research supervisor and DARA, conducted over four sequential prompts rather than a single research request. This sequencing is documented here because it shaped the paper's findings directly. 

The first question posed to DARA was broad and normatively loaded: does the introduction of AI kill the spirit of entrepreneurship, especially among the younger generation? DARA's initial response, built substantially on US-sourced data, distinguished entrepreneurial activity from entrepreneurial spirit and proposed a decoupling thesis rather than answering the question as posed. 

The second question was a direct methodological challenge from the supervisor, noting the US-centric character of the first response and requesting a global perspective. This produced a materially different picture, one in which the individual-psychology framing of the first question travels poorly outside a small number of AI-saturated, individualist labor markets. 

The third question sharpened the inquiry further, replacing the ambiguous "spirit" framing with a more falsifiable one: is AI democratizing entrepreneurship, or concentrating it among better-resourced actors? This question was posed by the supervisor as a refinement following review of the first two answers, and it produced the paper's clearest empirical finding, namely that democratization and concentration describe different layers of the same phenomenon rather than competing answers. 

The fourth question extended the inquiry to sectoral, occupational, and skill-level granularity, asking which industries, professions, and skills are most affected by AI in entrepreneurial terms. This required a corrective step, described below, that materially improved the evidentiary quality of the paper's final section. 

4.2 Source verification and a note on methodological discipline 

During research for the fourth question, DARA cited a working paper hosted on SSRN as a source for claims about job displacement and freelance labor markets. On direct request from the supervisor, DARA retrieved the paper's listing directly from SSRN rather than relying on secondary characterizations. This retrieval revealed the paper to be a self-posted, non-peer-reviewed synthesis report by an independent author with no institutional affiliation, presenting figures with a specificity (for example, exact job-loss counts to the ones digit) unsupported by any visible original methodology. The paper was excluded from the evidentiary base, along with several downstream claims that traced back to it via content-aggregation sites. 

This episode is documented here as a methodological illustration of the verification discipline applied throughout the research process, consistent with the broader DARA governance principle that sources reached indirectly must be checked against their primary listing before being treated as load-bearing evidence. 

4.3 Scope, division of labor, and limitations 

DARA's contribution to this paper is descriptive and synthetic: the identification, retrieval, and cross-referencing of publicly available data and research literature in response to the supervisor's questions. Interpretive judgments about which findings merit emphasis, the paper's normative implications, and all editorial decisions prior to publication remain the responsibility of the iNNOV8 Research Team. 

The evidence base draws predominantly on English-language sources and on organizations headquartered in North America and Europe (the World Economic Forum, the OECD, the World Bank, the US Census Bureau, and the Global Entrepreneurship Monitor), which may itself reproduce the geographic bias Section 2.2 attempts to correct for. A number of cited figures were retrieved via web search but not independently cross-checked against their primary institutional listings in the manner described in Section 1.2, and should be verified before this draft is finalized for external circulation. 

---

References 

Note on referencing: entries marked with a URL were retrieved and verified directly. Entries without a URL were identified during the research process but require primary-source verification before this draft is finalized. 

Brookings Institution. (2026). AI growth acceleration versus distributional fairness. https://www.brookings.edu/articles/ai-growth-acceleration-versus-distributional-fairness/ 

Global Entrepreneurship Monitor. (2025/2026). Global Report. [URL to be verified] 

Gusto. (2026). 2026 New Business Formation Report: How Artificial Intelligence Is Reshaping Who Starts a Business in America. https://gusto.com/resources/gusto-insights/new-business-formation-2026 

Guo, X., et al. (2025). Skill-biased technical change, again? Online gig platforms and local employment. [Cited via secondary review; primary source to be verified.] 

Hui, X., Qiao, D., et al. (2024). AI and Jobs: Has the Inflection Point Arrived? Evidence from an Online Labor Platform. arXiv:2312.04180. https://arxiv.org/pdf/2312.04180 

OECD. (2025). AI adoption by small and medium-sized enterprises. Discussion paper prepared for Canada's 2025 G7 Presidency. https://www.oecd.org/en/publications/2025/12/ai-adoption-by-small-and-medium-sized-enterprises_9c48eae6.html 

OECD. (2025). Emerging Divides in the Transition to Artificial Intelligence. OECD Regional Development Papers, No. 147. https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/06/emerging-divides-in-the-transition-to-artificial-intelligence_eeb5e120/7376c776-en.pdf 

Oxford Insights. (2024). Government AI Readiness Index. [URL to be verified] 

Stanford Digital Economy Lab. (2025). Study on AI-exposed occupation hiring trends. [Cited via secondary coverage; primary source to be verified.] 

Unnamed author(s). (2025). AI as "Co-founder": GenAI for Entrepreneurship. arXiv:2512.06506. https://arxiv.org/pdf/2512.06506 

World Bank. (2025/2026). Data on global distribution of AI models, startups, and venture capital. [Cited via Brookings, 2026; primary source to be verified.] 

World Economic Forum. (2025). Future of Jobs Report 2025. https://reports.weforum.org/docs/WEF_Future_of_Jobs_Report_2025.pdf 

Additional entries, including US Census Bureau Business Formation Statistics and Business Trends and Outlook Survey, SBA Office of Advocacy, Harris Poll/American Express survey, Boston Consulting Group 2026 executive survey, and Miller, D. (2020), cited via UNESCO entrepreneurship education literature, require primary-source retrieval before this reference list is finalized.

Related Researches