Firms built around AI are scaling knowledge work without the knowledge workers.
Listen to this article:
For several years now the idea that AI will transform organizations has been less a finding than a forecast repeated in keynotes, board decks, and earnings calls with a confidence that outpaced the evidence. The new working paper, “AI-Native Firms,” co-written by HBS AI Institute PI Rembrand Koning, offers one of the first empirical looks at what AI-enabled organizational transformation may look like in practice. Studying Y Combinator startups from 2020 to 2024, they ask straightforwardly whether organizations built around AI are actually organized differently. The answer turns out to be yes, but the why points somewhere most of the AI organizational transformation story has missed.
Key Insight: The AI-Native Firm Has a Distinct Shape
“Self-identification lets us rely on the founders’ beliefs about whether AI is central to their firm.” [1]
Y Combinator (YC) invites founders to publicly tag their own companies for discovery, including with an “AI” tag. The researchers use this as a starting point, then link those startups to PitchBook to measure financing and valuations, and to Revelio labs to measure organizational structure. To make fair comparisons, they compared AI-tagged startups with non-AI ones in the same industry and cohort, and extended the analysis beyond YC by training a model on YC startup descriptions to predict AI status among a broader set of U.S, venture-backed startups in Pitchbook.
Compared with non-AI startups, AI-native firms are about 25% smaller in the YC sample and 12% smaller in the broader PitchBook sample. They also employ a different mix of people: more engineers, fewer operations, finance, sales, and admin roles, fewer entry-level employees, and fewer managers, making their hierarchies flatter. For example, Educato AI, an AI exam-prep platform, runs on 7 employees while Careerist, which delivers courses through human tutors, has 912. Meanwhile, AI-native firms secure more funding and achieve higher valuations on a per-employee basis.
Key Insight: Two Channels, One Winner
“If this pattern holds, then building, importing, and orchestrating model capabilities may matter as much for competitive advantage as building human ones.” [2]
The authors distinguish between two ways that AI can reshape firms. The first is the process channel: when employees use AI tools like ChatGPT or Claude to do their existing work faster or better. In their job posting, AI-native firms name specific tools like Cursor or Claude at roughly 2.6 times the rate of non-AI peers, yet that difference fails to account for smaller headcounts or flatter structures.The stronger evidence comes from the product channel: building AI itself into what the firm sells, so that the work happens inside the product. To measure this channel, the authors classify YC startup descriptions based on whether AI automates work, augments experts, or provides AI infrastructure. They find that about two-thirds of AI-tagged YC startups embed AI directly into their products.
They highlight FazeShift, a company that builds AI agents for accounts receivable (AR), as an example of the radical possibility of the AI product channel. A conventional AR software company would likely need a large team to build custom workflows, support edge cases, and staff operational work, but FazeShift’s AI agents perform much of that work themselves, making the company smaller and more efficient because gaining more customers does not require an expansion of internal staff. This means that the core capabilities of a firm like FazeShift may increasingly be constructed through AI models, data, and product-embedded workflows, rather than a growing internal workforce.
Why This Matters
In a world with more competitors, faster product cycles, and lower barriers to entry, this research highlights that business leaders and executives can begin redefining success-with-AI in relation to value per employee. That starts with learning to scale through compute by embedding AI into products and services so more customer value can be delivered without a proportional expansion of headcount. It also means moving toward a more expert-dense workforce, where smaller teams build, supervise, and improve systems that handle day-to-day operations and maintenance themselves. Finally, leaders should reassess their firm hierarchies: high functioning firms are flatter with fewer layers, suggesting some coordination and routine knowledge work can move into tools and products. The firms that win may be those that stop treating AI as a productivity add-on and start treating it as a core part of their product.
Bonus
AI strategy is not just a technology rollout, but a redesign challenge. For another look at how fragmented workflows and legacy organizational design are impediments to AI ROI, check out Why Your AI Strategy May Be Failing.
References
[1] Kim, Hyunjin, and Rembrand Koning, “AI-Native Firms,” Harvard Business School Working Paper, No. 26-090 (June 2026): 9.
[2] Kim and Koning, “AI-Native Firms,” 4.
Meet the Authors
Hyunjin Kim is Assistant Professor of Strategy at INSEAD.
Rembrand Koning is Mary V. and Mark A. Stevens Associate Professor of Business Administration at Harvard Business School, and the co-director and co-founder of the Tech for All lab at the HBS AI Institute.
Watch a video version of the Insight Article here.