aibl Research: What Can We Learn From AI-Native Firms?

A new Harvard Business School working paper by Hyunjin Kim (INSEAD) and Rembrand Koning finds that AI-native firms are structurally different from their peers: leaner, flatter and more senior, and built so that AI does work inside the product rather than only helping staff work faster. The firms that look genuinely different are the ones that productised a capability they would otherwise have delivered through headcount.

Key takeaways

• AI-native startups run around 25% leaner than comparable non-AI startups, and about half the headcount three years in, while raising similar funding.
• They are flatter and more senior, with engineers a larger share of the team and fewer entry-level roles.
• The key divide is the process channel (AI as an internal tool) versus the product channel (AI built into what the company sells).
• Embedding AI into the product, not internal tool use, is what predicts smaller, flatter firms; about two-thirds of AI-native startups do this.
• The effect is largest in service businesses that used to scale by hiring, where AI-native firms run at roughly 30% of peers’ headcount.

We spent some time with the new HBS working paper by Hyunjin Kim (INSEAD) and Rembrand Koning, because it offers the clearest picture so far of how firms built around AI are organised. The essence is that the ‘AI natives’ do look structurally different, and the differences point to where AI value comes from.

The authors studied Y Combinator startups from 2020 to 2024, plus the wider universe of US venture-backed firms, linking each to workforce data on team size, seniority, and hierarchy. Then they compared those with the normies and identified four patterns.

How AI-native firms are built differently

They are smaller. AI-native firms run roughly 25% leaner than comparable non-AI startups. Three years after founding, they employ about half as many people. Yet they raise similar funding and hit comparable valuations, meaning they generate far more value per head.

They are flatter. AI-native firms carry about “half a seniority level less” in their hierarchy and 15% fewer managers, even after controlling for size. The subtext is that when AI is woven into productive capability, you need less of the coordination layer.

They have more experts per square foot. Engineers make up a larger share of the team as sales, finance, operations, and admin all shrink. These firms skew senior, with fewer entry-level workers, not more. The paper supports the case that AI reduces the need for junior staff, at this stage of development anyway.

Process channel versus product channel

The most useful idea in the paper may be the distinction between two ways firms use AI. The process channel is AI as an internal tool: workers using LLMs to move faster. The product channel is AI built into what the company sells, so the software itself does work that used to require people, or can offer a new product/service that was previously impossible or non-viable at scale.

The organisational differences they identified trace almost entirely to the product channel. About two-thirds of AI-native startups embed AI directly into their products. When the authors measured internal tool use through job postings, it did not predict smaller or flatter firms, but embedding AI into the product did.

This matters because it reframes what “adopting AI” means. Handing staff better tools helps at the task level, but doesn’t reshape the organisation. The firms that look genuinely different are the ones that productised a capability they would otherwise have delivered through headcount (or not at all). Gamma reached tens of millions of users and $50m in revenue with about 30 people by turning each slide-deck request into a product interaction rather than an internal workflow. FazeShift automates the full accounts-receivable cycle with a team of ten.

Where the impact is biggest

The effect is largest in service businesses that historically scaled by hiring: therapy, tutoring, exam prep. In these industries, AI-native firms run at roughly 30% of the headcount of non-AI peers. One AI exam-prep platform employs seven people against 912 at a non-AI competitor delivering courses through human tutors.

Two cautions

Two cautions belong alongside the findings. The design is observational and descriptive, so it establishes patterns not causes, and these are young firms whose structures may shift as they scale. There is also a distributional question: AI-native workforces are more concentrated in places like San Francisco, more male, and drawn from more elite institutions. Smaller firms need not mean fewer jobs overall if AI lets many more companies launch, but who gets those jobs?

What this means for mid-market leaders

The lesson is that the competitive edge from AI may come less from equipping people with tools and more from re-engineering the product so the technology does the work directly. That is a harder change, with a bigger payoff.