AI-native startups are building smaller, flatter teams that skip the traditional entry-level rung entirely. A working paper from Harvard Business School and INSEAD, titled “AI-Native Firms,” examined Y Combinator startups from 2020 to 2024 and a broader set of US venture-backed startups from the same period, Business Insider reported.
The numbers are specific. AI-native startups are 25% smaller than their non-AI peers. They employ about 13% more engineers. Their shares of entry-level workers and managers are each roughly 15% lower. The share of senior workers is 20% higher.
What “AI-Native” Means
The paper defines AI-native firms through two productivity shifts, according to Business Insider. The process channel: using AI internally to make employees faster at coding, selling, designing, and coordinating. The product channel: embedding AI directly into what the company sells, so customers use the product to perform work that previously required human teams.
Companies that do both build fundamentally different organizations. Fewer layers of management. Fewer junior employees learning on the job. More senior engineers who already know how to direct AI tools effectively.
The Concentration Problem
The researchers found a demographic pattern that cuts against the narrative of AI as a democratizing force. “AI-tagged firms employ smaller teams of more talented and technical workers,” the authors wrote, per Business Insider. “These workers are especially likely to be graduates from elite institutions, concentrated in Silicon Valley, and male.”
The concern is compounding: if AI tools accelerate learning for those who use them, and adoption rates vary by demographic group, the performance gaps widen over time. The researchers wrote that “differential adoption rates may translate into widening performance gaps,” both for individual workers within firms and for the entrepreneurs who found them.
What This Changes for Agent-First Teams
For founders building with AI agents, the study quantifies something that has been anecdotal until now. The vibecoding trend has made it easier for non-engineers to prototype, but the firms that actually ship and scale are concentrating on senior technical talent who can direct AI tools, not junior workers who might be partially replaced by them.
The entry-level career ladder in software, which has served as the primary pipeline for training the next generation of engineers, is compressing. If AI-native startups are the template for the next decade of venture-backed companies, the question is where junior developers will get their initial experience if fewer companies are willing to invest in training them.