Sometime in Q3 2025 I sat across a table in Bangalore from a founder I respect. He was raising an $8 million round. The team was four people. Two engineers, one half-time designer, and him. The pitch deck listed no “Head of” anything. The revenue multiple on the deck was large enough that a 2015-era VC would have asked where the rest of the team was. The VC across the table in 2025 asked the opposite question: why do you need four?
He closed in six weeks.
Sometime in Q1 2025 I made the opposite decision myself. I evaluated — seriously — a physical-hardware vertical for one of my ventures. The unit economics were compelling. The market was real. I ran the head-count math twice and walked away. The vertical needed a compliance officer, a regulatory lead, a hardware QA team, and a supply chain hire before it could ship anything resembling a first product. The math said seven to nine full-time humans, minimum, before month one of revenue. That number broke my operating model. I killed the vertical on a Wednesday and did not reopen the folder.
Both stories are PRED-006 in action. One where the prediction holds. One where it does not. This issue is about where each one applies.
PRED-006 — By December 2029, 20% of Series A rounds — $5M–$20M rounds from top-tier VC firms — will go to companies with fewer than five full-time human employees. Humans serve as strategic decision-makers and agent orchestrators, not execution labour.
Confidence: 4 out of 5.
The alignment between VC incentives and agent-first economics is the strongest alignment in the category.
Recent Y Combinator batches show an increasing number of solo and duo-founder companies. Garry Tan stated publicly in March 2025 that about 80% of the Winter 2025 batch was AI-focused, with several companies reaching as much as $10 million in revenue on teams of fewer than ten people. VCs have started writing “revenue per employee” on the front of their diligence memos — and AI-native companies show 10 to 50 times higher revenue per employee than traditional SaaS. The one-person conglomerate model (last issue’s subject) demonstrates that a single person can run multiple agent-powered ventures with real revenue. The Series A question was whether VCs would fund the model at scale. The answer in 2025 and 2026 is already: they are.
That is the pro-case, and the chapter makes it well. Every piece I read on this subject makes the same pro-case. What most of those pieces skip — and what I want to put on the record here — is the counter-case. Because PRED-006 is more defensible if I name its limits.
Here are the verticals where the small-team Series A does not apply, and where the prediction should not be read as a general claim:
Hardware. Physical products require hardware engineers, mechanical engineers, supply chain, QA, and regulatory roles that do not compress with agents. Five people cannot ship a device at Series A scale. I know because I looked at it in 2025.
Deeptech with physical lab requirements. Materials, biotech, fusion, advanced manufacturing. Lab operations are not agent-automatable at current state.
Heavily regulated medical. FDA, EMA, and equivalents require named responsible persons, qualified persons, and quality-management headcount written into the approval pathway. You cannot substitute an agent for a person who signs a regulatory filing.
Defense. Clearance requirements, supply-chain security obligations, and procurement norms impose a human-headcount floor by contract.
Anything with a compliance headcount floor written into a government procurement requirement. India’s CERT-In, the US federal StateRAMP/FedRAMP paths, EU DORA — each names human roles that must exist. The floor is legal, not operational.
PRED-006 is specifically about software Series A. Agent-first, digital-delivery, globally distributable software. In that arena the prediction is strong. Outside that arena it quietly does not apply, and naming the exceptions is how I keep the prediction honest.
That is the practitioner receipt: my own 2025 decision to kill a hardware-adjacent vertical because the headcount floor would have broken the conglomerate model. A prediction I believe in strongly enough to name, and one I respect enough not to over-claim on territory it does not cover.
The published falsification trigger:
If by December 2029, fewer than 10% of Series A deals from top-20 VC firms go to companies with fewer than five full-time employees, or if no major VC publicly endorses the agent-first company model, this prediction is wrong.
That is the stake as written. Here is the way I am actually most likely to lose it.
Definitions. If “Series A” drifts to mean $15M–$40M rounds by 2029 — because the median Series A keeps creeping up — the 20% number applies to a different denominator than the one I had in mind. And if the top-tier VC firms keep writing seed-stage tickets that sit between $5M and $20M but do not call them Series A, the small-team rounds live in a category my prediction did not nominate. That is how this one loses without the underlying thesis being wrong. I will watch the round-label drift as carefully as the team-size distribution.
I want two specific pieces of data.
If you are inside a VC fund: share one 2025 or 2026 Series A deal where the team was fewer than five and the vertical was outside software — specifically hardware, deeptech, regulated medical, or defense. Anonymise the founder, not the vertical. If multiple such deals surface, the exception list above is wrong and I should strengthen the prediction, not caveat it.
If you are a software founder who closed a Series A in 2025 or 2026 with fewer than five employees: send me the team size, round size, and vertical. I will publish an aggregate distribution at atin-agarwal.com/predictions/pred-006-small-team-series-a/. The real test of PRED-006 is the trend line between now and 2029, and the only honest way to draw it is with actual deals.
If you are a founder in a software vertical: the small-team Series A window is open. Revenue per employee is the metric that gets you funded now. Build the agent loop, measure the ratio, put it on slide two of your deck.
If you are a founder in a vertical on the exception list: stop reading small-team VC Twitter. It is not your benchmark. Size your team for the category you are actually in — hardware, deeptech, regulated medical, or defense has a headcount floor and the floor is not embarrassing. It is accurate.
If you are a VC: your pattern-match from software small-teams does not transfer to deeptech diligence. Two different playbooks. Using one for the other is how you miss the deeptech deal and overpay for the software one.
If you are an engineer considering joining a small-team Series A company: the “review and correct agent output” seat is the one you want. It is the seat that survives, and in a four-person company it is the seat with the most impact. Ask for it explicitly in the offer.
This issue is drawn from Chapter 9 of The AI Agent Economy — 15 falsifiable predictions with dates, numbers, and explicit triggers for being proven wrong. Pre-order on Kindle — $9.99. Release July 1, 2026. atin-agarwal.com/books
Read the full PRED-006 entry on the public tracking page → atin-agarwal.com/predictions/pred-006-small-team-series-a/
Previous issue: Issue 05 — 10,000 one-person conglomerates by 2028, and the metric that tells you if you’re really one
Next issue: Issue 07 — A SaaS feature will cost 1/10th to ship, if you pay the quality tax (publishing May 24)
The hardware vertical decision is my own 2025 operating decision; the $8M round story is composite and anonymised, shared with founder’s permission.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.