The research, and why it's more credible than the usual claims
Most "AI is shrinking teams" claims come from vendor surveys with an obvious incentive. A June 2026 working paper from Harvard Business School and INSEAD researchers Hyunjin Kim and Rembrandt Koning stands apart: using Y Combinator batches from W20 through F24 and US venture-backed startups first financed between 2020 and 2024, it found that startups embedding AI directly into their products employ roughly 25% fewer workers than comparable non-AI startups in the same industry cohort, while reaching comparable valuations. Same outcome, fewer people, which is a meaningfully different claim than "AI replaces jobs" in the abstract.
It's not just fewer people, it's a different shape of team
The same paper found AI-native firms carry a higher engineer share and fewer entry-level workers and managers relative to their non-AI peers. A separate data point on go-to-market specifically found companies under $25 million ARR with high AI adoption running with just 13 GTM full-time employees versus 21 for traditional SaaS peers at similar scale. The pattern across both engineering and GTM is the same: fewer layers of coordination and fewer roles whose job was primarily to do repeatable work that AI tooling now handles, concentrated instead around a smaller group making the actual decisions.
The honest caveat the researchers themselves raise
Kim and Koning explicitly caution against reading their 25% figure as a forecast for the broader job market, and other analysis adds useful skepticism: one researcher points out that the average seed-stage consumer startup in 2024 already had fewer than 3.5 employees, down from 6.4 in 2022, a shift that predates most current AI tooling and may reflect general efficiency trends as much as AI specifically. Less rigorous reports claiming startups run with "two-thirds fewer employees" come from self-reported, unmethodologically-transparent surveys and shouldn't be weighted the same as the HBS-INSEAD paper. The honest takeaway is the 25% figure, grounded in real data, not the more dramatic unverified claims circulating alongside it.
Why this is the model a studio runs on by design
This is precisely the structural bet a venture studio makes on purpose, not by accident. A small, senior, AI-native team that reuses infrastructure, design systems, and hard-won market lessons across every product it builds doesn't need a large headcount to reach the valuations larger teams chase, it needs the right five or six people making good decisions with AI doing the repeatable work underneath them. That's not a cost-cutting story, it's the same operating logic the Kim-Koning data describes at the company level, applied deliberately to how a studio originates and builds its own ventures, and why the studio model compounds, each product gets faster and cheaper to build than the last.