AI-native startups run with 25% fewer people than comparable non-AI startups at the same stage, according to a working paper from Harvard Business School and INSEAD (HBS Working Paper 26-090), but they are not simply doing more with less across the board. The paper, by Hyunjin Kim of INSEAD and Rembrand Koning of HBS, studied Y Combinator batches from W20 through F24 alongside U.S. venture-backed startups founded between 2020 and 2024. It found these companies carry a 13% greater share of engineers than their peers, while entry-level workers and managers each make up roughly 15% less of headcount, and the org chart itself runs about half a seniority level flatter.
That is a specific shape, not just a smaller one: more engineers, fewer juniors, fewer managers, less hierarchy between the founder and the person writing code.
The senior-only playbook
Replika and Wabi founder Eugenia Kuyda is building to that shape deliberately. She has said AI made junior hires "extremely expensive and completely unsustainable for a startup," because every new hire now competes against what she calls a "1,000x engineer" (Platformer). Her plan for Wabi is 10 to 15 senior people, each holding meaningful equity, backed by contractors for the rest of the work, rather than a conventional junior-heavy engineering department.
Anthropic is describing a version of the same logic at a much larger scale. Co-founder Jack Clark said the company is "hiring more people with lots and lots of experience than we did before, because the returns on intuition are much greater than before" (Reason). His reasoning: Claude now runs the experiments that used to require an engineering team to execute, so Anthropic can add senior researchers without scaling the engineering headcount that used to surround them.
The data complicates the story
Set against that, SignalFire's State of Talent Report 2026 found engineering hiring is one of the more resilient categories in the market, not a shrinking one. Engineers made up 55% of all new hires in 2025 at the 12 companies SignalFire classifies as "Tech Majors," up from 46% in 2019. Total hiring at those companies fell 25% against 2019 levels, but engineering-role hiring fell only 11%, and early-stage startups collectively hired 7% more engineers in 2025 than in 2019 (TechCrunch).
SignalFire's head of research, Asher Bantock, put the gap between the narrative and the numbers directly:
"The rationale given for lots of layoffs is consistently AI, and specifically they'll say AI with respect to code; they'll say one engineer could do the job of however many engineers in the past," but "what we're seeing on the ground is a little inconsistent with that." (TechCrunch)
Two more data points cut against a blanket junior-to-senior swap. AWS CEO Matt Garman said Amazon will hire 11,000 interns and recent graduates in 2026, and pushed back on mass entry-level job loss forecasts: "If you believe that half of jobs get wiped out, the whole economy collapses on itself" (Fortune). The Pragmatic Engineer newsletter found Google advertised 62% more engineering roles than the year before, even as Meta, Oracle, and TikTok posted fewer openings, with the U.S. and U.K. the only tracked countries where developer job vacancies rose year over year (The Pragmatic Engineer).
What to do with this before you write the next job post
- Both patterns above are true at once, at different layers of the same org. Engineering headcount is holding up or growing at many companies, early-stage included.
- What is shrinking is the middle: entry-level slots and the manager layer that used to sit between a founder and an individual contributor, the same 15% the HBS paper measured.
- That is the layer Kuyda and Clark are both describing when they talk about paying for judgment over hours worked.
Founders should also weigh the cost of overcorrecting on the AI side of that equation. Duolingo built its 2025 "AI-first" push around factoring AI usage into employee performance reviews, then CEO Luis von Ahn dropped that policy (Fortune). A mandate that scores employees on AI tool usage is a different thing from hiring senior engineers who already use AI well, and one company that tried the former has already walked it back.
The action for this week: before cutting a junior req, check whether you are actually removing a management layer or removing your future senior bench. The HBS data says AI-native firms run flatter and more senior, not headcount-free, and SignalFire's data says engineering hiring broadly has not contracted the way layoff announcements suggest. Cut the manager layer if the org supports it. Keep a narrow junior pipeline running anyway, because the senior engineers you will need to hire in three years do not exist yet if nobody trains them now.