GenAI as a Co-Founder: How Tiny Teams Sparked a Startup Boom After ChatGPT
Introduction: entrepreneurship meets a digital co-pilot
If you’ve ever dreamed of starting a business but felt held back by not having a big team or mountains of cash, GenAI might just feel like a game changer. The rise of generative AI (GenAI) tools—think ChatGPT and friends—has pushed a big question to the forefront: can AI actually help people start new firms, not just run existing ones? Researchers Cai, Gu, Sheng, Xia, Zhao, and Zhu set out to answer this, using a natural experiment sparked by ChatGPT’s global debut in November 2022.
What they found in China is striking: in places that already had more AI know-how before ChatGPT, new firms popped up faster after the tool’s release. And the boost wasn’t spread evenly across all firms. It was largely about small, lean startups—tiny teams with light capital and little prior entrepreneurial experience. In other words, GenAI acted like a “digital co-founder” that lowers the barriers to entry and reshapes how firms are formed from the ground up.
What does GenAI as a co-founder really mean?
Traditional entrepreneurship often means recruiting a team with specialized skill sets: software developers, designers, marketers, legal help, and financing you can draw on as you grow. GenAI changes that math. With a universal language interface and a broad range of tasks GenAI can perform—coding, writing, data processing, design, marketing content, even some strategic thinking—a single founder or a tiny team can prototype, test, and ship products much more quickly and cheaply.
The authors push a simple, concrete idea: GenAI can substitute for some of the labor, management, and financial frictions that typically slow down or block the start-up process. When GenAI is readily available in a founder’s toolkit, the “minimum viable team” can be a lot smaller. That’s why the paper emphasizes the possibility of GenAI serving as a “co-founder” that complements human founders rather than replaces them outright.
The data: a high-resolution, grid-by-grid view of China’s startup landscape
To test these ideas, the researchers combine two large, nationwide datasets from China:
- Firm registrations (2021–2024): more than 12 million newly established firms, with details like registered capital, ownership, and founding team.
- AI patent data (2010–2019): about 340,000 AI invention patents, used as a proxy for local AI-specific human capital and know-how before GenAI arrived.
They map every new firm and every AI patent to a hexagonal grid cell about 5 square kilometers in size, creating a fine-grained grid-by-quarter panel across the country (166,156 cells in total). Then they classify each grid by its “AI exposure” before 2020: grids with at least one AI patent are labeled high-AI exposure; others are low-AI exposure.
A sharp, plausible shock: ChatGPT’s release in 2022 provides the exogenous event
One of the clever parts of the study is using ChatGPT’s global release as a near-synchronous shock to GenAI diffusion. Because the diffusion of this technology happened quickly and widely, the authors can compare startup activity before and after the ChatGPT moment, within the same city, across neighboring grids that differed in pre-existing AI capital.
In plain terms: within a city, if two neighboring grids had different levels of AI know-how before 2020, how did their startup numbers change after ChatGPT’s debut? If GenAI truly matters, the grids with more AI head start should show a bigger bump in new firms after ChatGPT, compared with their nearby peers.
Small firms rise, large firms retreat
Here’s the headline result, in accessible terms:
- Overall, high-AI grids saw a noticeable rise in new firm formation after ChatGPT’s release. On average, these grids added roughly five more new firms per grid per quarter than their low-AI peers.
- When you scale this up, the nationwide impact translates to about 51,000 extra new firms per quarter across all high-AI grids, roughly 6% of total firm entries in the period after ChatGPT rolled out.
- The effect is asymmetric by firm size: it’s driven entirely by small firms. Large-firm entry actually declines in high-AI grids after ChatGPT.
Why this pattern? GenAI lowers fixed costs and the need for large, specialized teams at the outset. Small, lean ventures can plug GenAI tools into core tasks—coding, marketing, customer support, product iteration—without hiring a big crew or pulling in large sums of external capital.
Mechanisms in plain language: three big levers GenAI lowers
1) Experience constraints (the “know-how” barrier)
- Before GenAI, first-time founders and serial entrepreneurs alike relied on experience to navigate startup tasks—drafting business plans, building products, handling marketing, and more.
- GenAI acts like a mentor and doer all in one. It helps inexperienced founders perform tasks that used to require seasoned labor, so first-time founders can move faster and more confidently.
2) Financing friction (the “who’s in” barrier)
- Startups often need multiple shareholders or investors to secure enough capital to launch.
- GenAI reduces the need for large co-founders or a big pool of capital at the entry point. Founders can still move fast with fewer investors, which is especially valuable for small ventures.
3) Labor intensity (the “how many people” barrier)
- Early-stage startups usually need a handful of people to run multiple functions.
- GenAI can handle many tasks across design, coding, content creation, data analysis, and basic operations, enabling a much smaller founding team.
But the effect isn’t uniform across all sectors
The study finds that the biggest bumps come in AI-adopting, downstream sectors—areas where GenAI tools can be embedded directly into products and services, such as retail, business services, and digital platforms. In contrast, traditional, capital-intensive sectors like construction or heavy manufacturing saw little or even negative effects on new firm formation.
This pattern makes sense: sectors that already lean on knowledge work, marketing, and digital services benefit most from AI-powered productivity boosts and fast prototyping.
Further evidence on the channels: what changes at the founding level?
Several deeper patterns emerge when researchers look at the founding team composition and founder experience:
- Serial entrepreneurs vs first-time founders: After ChatGPT, areas with more AI know-how saw more first-time founders entering. Serial entrepreneurs’ participation declined in high-AI zones, hinting that GenAI helps novices enter the market more easily.
- Size and ownership: New firms in AI-rich grids tend to have fewer shareholders and smaller founding teams. Ownership shares still skew toward individuals, but the average number of shareholders drops, signaling lighter initial financing and coordination costs.
- Founding team composition: Executive teams at entry shrink as GenAI tools substitute for early-stage managerial labor. This trend is especially pronounced for small firms.
In other words, GenAI nudges the startup world toward leaner, faster, low-cost formations—without necessarily reducing the overall appetite for entrepreneurship. It’s not that founders stop partnering or raising capital; it’s that the required scale and staffing at the outset can be smaller.
Robustness and credibility: making sure the effect isn’t a mirage
The researchers run a robust battery of placebo and robustness tests to ensure the results aren’t just capturing general regional vigor or some other trend. A few highlights:
- Non-AI patents as a placebo: Replacing AI patents with non-AI patents largely wipes out the post-chat effects, suggesting the results hinge specifically on AI-related human capital, not on broader innovation.
- Residuals-based placebo: They pare down pre-2019 entrepreneurial activity by removing the portion explained by AI patents. The post-ChatGPT effects shrink substantially, reinforcing that AI-specific human capital is the key driver.
- Randomized label placebo: They shuffle the “high-AI” label across grids and re-run the analysis 100 times. The placebo results cluster around zero, indicating the observed effects aren’t just a statistical fluke.
- Geography and sample robustness: Excluding top-tier provinces or using AI-matched controls within cities still yields strong, consistent results. They also test different definitions of “small firm” (various capital thresholds) and still see the same pattern: big lift for small firms, with large firms receding.
Who benefits most? Industry and AI-exposure nuances
Beyond the broad small-vs-large story, the paper digs into heterogeneity across industries and AI-relevance dimensions:
- Industry variance: The strongest post-ChatGPT responses are in Retail, Business Services, and Technology Promotion/Applications sectors. Other service-oriented and consumer-facing areas also show positive effects, while traditional high-capital industries see muted or negative responses.
- AI exposure dimensions: They create AI-relevance scores for firms (upstream AI development, downstream AI applications, and entrepreneurship usefulness). The lift in new entries is larger for firms with high downstream AI relevance and those where GenAI can directly support entrepreneurial tasks (think no-code tools, marketing automation, product onboarding, etc.). Upstream AI-resemblance (things like hardware, data centers, foundational AI tech) shows smaller effects—these areas remain more capital- and expertise-intensive.
This pattern lines up with a practical takeaway: GenAI adds the most value where it’s easiest to plug into existing business processes and customer-facing activities, rather than where you have to build new AI technologies from the ground up.
A note on the “co-founder” metaphor
The authors aren’t saying AI will replace human founders. Instead, GenAI acts as a co-pilot or co-founder in the sense that it shares the cognitive load of early startup work—the brainstorming, drafting, coding, marketing, even some strategic planning. The result is a more democratized entrepreneurial landscape: people with fewer resources or less prior startup experience can still form viable, lean ventures when GenAI is at their side.
Real-world implications: what this could mean for founders, educators, and policymakers
- For aspiring entrepreneurs: GenAI tools can materially shorten the path to launching a business. If you’re in a city or region with AI know-how, your odds of starting a small, lean venture rise, because the tools are better aligned with your local strengths and talent pool.
- For regional development: The benefits cluster in areas with more AI-related human capital. This suggests that building local AI literacy, AI-related training, and access to AI-enabled workflows could boost entrepreneurship in more parts of the country or economy.
- For incumbents and policy: The shift toward lean early-stage firms doesn't mean big firms lose. It suggests a dynamic where startup activity is more vibrant at the margins, potentially promoting competition and faster product iteration across industries.
Practical takeaways and ideas for action
- If you’re a founder or would-be founder: Consider how GenAI tools can compress your minimum viable product timeline. Focus on tasks that are transformation-ready for AI assistance (content generation, outreach, basic product development, data analysis). Start lean, then scale with a clear plan to validate product-market fit.
- If you’re an educator or mentor: Emphasize AI literacy as a core founder skill set. Teach practical workflows that integrate GenAI for prototyping, market research, customer support, and marketing—not just as a cool gadget, but as a practical toolkit.
- If you’re a policymaker or city planner: Invest in AI-related human capital in your region—training programs, incubators, access to compute resources, and partnerships with AI-focused industries. The study shows a tangible link between local AI know-how and startup formation, especially for small, agile ventures.
- If you’re an investor or corporate innovator: Look for startups that are using GenAI to reduce founding frictions—teams that can leverage AI to prototype quickly, test ideas with real customers, and operate with small but capable teams.
Limitations and avenues for future exploration
No study is perfect, and the authors acknowledge several caveats. The setting is China, with its own regulatory and policy environment, which may differ from other countries. The measure of AI exposure relies on pre-2020 AI patents as a proxy for local AI human capital; while robust, it’s not a perfect capture of capabilities. The long-run effects—on job growth, productivity, and industry structure—are important questions for future research, particularly as GenAI tools evolve and become even more integrated into everyday business practice.
Key takeaways
- GenAI can function as a digital co-founder, dramatically lowering the barriers to startup entry for small, lean ventures.
- The post-ChatGPT surge in firm formation was concentrated in grids with higher AI-specific human capital and was driven entirely by small firms, with large-firm entry actually declining.
- The mechanism is largely about reducing fixed costs and the managerial labor required to launch a firm: fewer shareholders, smaller founding teams, and less serial founder activity, especially among first-time entrepreneurs.
- The effects are strongest in AI-adopting, downstream industries where GenAI tools can be quickly integrated into products and services.
- Robust placebo and robustness tests strongly support the interpretation that AI-specific human capital—rather than general innovation—drives the observed boost in entrepreneurship.
- The findings imply a broader democratization of entrepreneurship: GenAI tools help more people start firms, particularly where local AI know-how exists, and reshape how ventures are organized from the outset.
- For practitioners and policymakers, the key implication is to invest in AI literacy and complementary skills at the regional level to unlock entrepreneurial potential and maintain competitive dynamics in the economy.
If you’re intrigued by the idea of GenAI as a co-founder, you’re not alone. The study paints a picture of a future where savvy individuals with modest resources can launch meaningful ventures with the help of AI, turning a once formidable barrier into a more navigable doorway. As GenAI tools keep improving, that doorway may only get easier to walk through.
Key Takeaways (condensed)
- GenAI lowers startup barriers, enabling lean, founder-light ventures to form more rapidly.
- After ChatGPT’s release, small firm entry rose in AI-rich areas, while large-firm entry fell, suggesting a shift toward lean entrepreneurship.
- The effect is strongest in AI-downstream, application-focused sectors; upstream, capital-intensive sectors show weaker responses.
- Founders in AI-rich grids showed fewer serial entrepreneurs, smaller teams, and fewer shareholders—clear signs of labor and financing frictions being alleviated by GenAI.
- Robustness checks (placebo, residualization, matching) support a causal interpretation tied to AI-specific human capital.
- Policy and education efforts that build local AI skills and infrastructure could amplify these entrepreneurial gains.
If you’d like, I can tailor this post to a specific audience (founders, students, policymakers, investors) or convert it into a shorter version for social media with additional practical prompts for implementing GenAI in a startup context.
