Generative AI Fuels Solo Entrepreneurship, but Teams Still Lead at the Top

Abstract: Recent advances in generative AI are reshaping who enters entrepreneurship, but not who reaches the top of the quality distribution. Using data on over 160,000 product launches on Product Hunt, we find that entrepreneurial entry increased sharply following the public release of ChatGPT-3.5, driven disproportionately by solo entrepreneurs. This shift toward solo entry is particularly pronounced in categories that historically favored team-based ventures. However, much of this growth reflects low-commitment, experimental entry and does not translate into greater representation among the highest-quality outcomes. Team-based ventures are increasingly dominant in the top tiers of platform rankings. These findings suggest that generative AI lowers barriers to solo entrepreneurship while reinforcing team-based advantages.

• The paper finds that generative AI expanded entrepreneurial entry, with solo launches growing nearly 90% more than team launches by 2025 and rising sharply in historically team-heavy categories.
• The paper shows that marginal solo ventures were more experimental and short-lived, as the one-shot entry rate increased from 95.1% to 97% while product distinctiveness also rose after ChatGPT’s release.
• The paper finds that teams retained an advantage at the top, with team representation in Product Hunt’s Top 10 increasing from 49.7% to 53%, particularly among teams of four or more members.

Overview

This paper examines how generative AI reshapes both the composition and quality distribution of entrepreneurial entry. Using 160,143 product launches on Product Hunt between April 2020 and July 2025, the authors exploit the public release of ChatGPT-3.5 in November 2022 as a quasi-experimental shock to access to generative AI, and estimate difference-in-differences (DiD) and event-study models comparing solo founders (one maker) with team-based ventures (two or more makers). The central finding is a divergence between quantity and quality: entry expands sharply, driven disproportionately by solo founders, but team-based ventures become more concentrated in the top tiers of platform rankings after the shock. The authors interpret this as generative AI lowering entry barriers while leaving the organizational requirements for top-tier outcomes intact, effectively relocating bottlenecks from venture creation toward sustained development and refinement.

Empirical setting and identification

The setting exploits Product Hunt's granular records of launch timing, founding team size at entry, category tags, post-launch updates, and daily rankings derived from platform-weighted upvotes and engagement. This allows joint observation of entry and early market evaluation at the moment of public market exposure, rather than relying on firm registrations or VC funding, which capture only a selective, later-stage subset of ventures. The authors define venture creation as first public market-facing realization, consistent with process models of venture creation.

The DiD specification compares log monthly entry counts for solo versus team ventures with founding-structure and time fixed effects; the event-study version traces dynamics relative to 2022. Identification rests on parallel trends between solo and team entry absent the ChatGPT release. The event-study estimates show small, statistically insignificant coefficients for event years $-3$ and $-2$, but the year immediately preceding treatment (2021) shows a significant negative coefficient ($-0.105$, $p = 0.006$). The authors attribute this to localized fluctuations in February, March, and August 2021; excluding those months attenuates the estimate to $-0.069$ ($p = 0.053$). This is a candid concession of an imperfect pre-trend, and readers should weigh it when interpreting the magnitude of the post-shock effect. Placebo permutation tests for entry and distinctiveness support that the results are not artifacts of arbitrary patterns.

Entry effects and cross-category heterogeneity

Total entry rose roughly 40% from 29,326 launches in 2022 to 41,169 in 2023, approaching twice the pre-release level by 2025. The increase is disproportionately solo: by 2025, solo entry grew nearly 90% more than team entry ($\beta = 0.621$, $p < 0.001$ in the event-study specification; the static DiD estimate is $\beta = 0.264$). The implication is that generative AI functionally substitutes for co-founders in enabling initial product creation.

The cross-category analysis shows the solo shift is strongest where it was previously hardest. Categories that were historically team-heavy—such as SaaS and Fintech—exhibit larger post-shock increases in solo-founder share, with a negative cross-sectional relationship between pre-shock solo share and post-shock change ($\beta = -0.324$, $-2$0). Quarterly series for representative team-heavy and solo-heavy categories show stable pre-treatment trajectories followed by divergence after Q4 2022, and results are robust to using the top 50, 150, or 200 categories by launch frequency. This pattern suggests generative AI partially substitutes for the coordination and capability previously requiring multiple human founders, at least at the entry stage.

Solo entry as low-commitment experimentation

A key contribution is characterizing the nature of the marginal solo entry. The paper measures "one-shot" entry—no product update within 12 months of launch (a window justified by a mean time-to-first-update of 9.6 months). The solo one-shot rate rises from 95.1% to 97% post-shock, more than double the increase observed for teams, translating to approximately 215 additional one-shot entries per year. Because the platform baseline one-shot rate is already near 95%, the authors argue this increment is substantively meaningful, though this framing also reveals that most entries on the platform—solo or team—are short-lived.

Solo ventures also show a differential increase in product distinctiveness, measured as the rarity of a product's tag combination relative to same-day launches ($-2$1, $-2$2; event-study coefficients of 0.028–0.037 in post-treatment years, with clean pre-trends). The authors interpret the combination of rising one-shot rates and rising distinctiveness as expanded recombinant search: generative AI enables solo entrepreneurs to explore less familiar regions of the product space through low-commitment, trial-and-error experimentation, consistent with real-options logic and with evidence that recombinant search yields more variable, lower-average-value outcomes.

Quality outcomes favor teams

The quality analysis uses Product Hunt's daily rankings as a proxy for early-stage market evaluation. The central result contradicts any simple democratization narrative: team share within the Top 10 increased from 50% (49.7%) to 53% ($-2$3) after ChatGPT's release, and rose from 38.1% to 39.5% in the Top 30 ($-2$4). As the tier definition broadens, the effect reverses—team share in the Top 50 is flat, and across all products it fell 15.1 percentage points (from 34.2% to 19.1%), mechanically reflecting the solo-driven surge in total entry.

Within the Top 10, gains concentrate among larger teams: teams with five or more members increase their share by 2.0 percentage points ($-2$5) and four-member teams by 1.1 points ($-2$6), while solo founders decline by 3.8 points ($-2$7) and duos are flat. The authors interpret this as evidence that top-tier outcomes depend on integrating tacit, experiential knowledge distributed across individuals—capabilities that standardized, codified AI outputs do not fully substitute, a point reinforced by evidence that different LLMs produce highly similar responses to open-ended prompts. The implication is an asymmetry: generative AI relaxes entry constraints but not the organizational requirements for reaching the upper tail of the quality distribution.

Limitations and open questions

The paper acknowledges that its evidence comes from a single digital platform focused on software launches; many entries are not incorporated firms and some are small or short-lived initiatives. Generalization to capital-intensive industries, incorporated firms, or longer-run outcomes such as survival, growth, and patenting remains untested. The ranking-based quality measure aggregates early community votes and engagement rather than realized economic performance, so the persistence of team advantages into later venture stages is an open question. The identification strategy also depends on Product Hunt-level composition not shifting for reasons unrelated to generative AI, and the significant negative 2021 pre-trend coefficient—though the authors show it is driven by three months—warrants continued attention. Finally, the authors note that distinctiveness gains at entry fade over time, raising the unresolved question of whether shared generative tools and inputs converge entrepreneurs toward similar solution spaces and reduce collective novelty.

Conclusion

The paper provides large-scale empirical evidence that generative AI expanded entrepreneurial entry—particularly solo entry into historically team-heavy domains—while concentrating top-ranked outcomes further among larger teams. The resulting quantity–quality divergence implies that entry-reducing policies alone will not generate high-growth ventures, and that generative AI reallocates rather than removes constraints in venture creation, shifting bottlenecks toward sustained development, coordination, and the integration of diverse human expertise alongside AI.