A Very Large Number of Very Small Teams

The intuitive read on a productivity shock is that it kills jobs: do the same work with fewer people, fire the rest. That read is half right and badly framed. Productivity does shrink the headcount needed to ship any given thing. What it does not do is hold the number of things constant. When the team-per-task collapses, the binding constraint stops being “how do I staff this?” and becomes “what else could I build?” The output of that question is not unemployment. It’s an explosion of new teams — what Naval, on a recent episode, called “a very large number of very small teams.”

That phrase is the seed for this post, but the claim only earns its keep if the numbers move. They do — though not always as cleanly as the hype suggests. Here’s the honest version.

The team is shrinking and the count is rising #

Start with formation. New US business applications have been running near historic highs — on the order of half a million filings a month — and roughly 60% of new business owners report using AI to launch. (Gusto) That’s the “multiplies the number of things built” half of the thesis, showing up in the registry data: more shots on goal, not fewer.

Now the team size. Firms founded in 2024 in AI-enabled industries carried about 6% fewer employees at the twelve-month mark than the 2023 cohort, and in the Bay Area the reduction was closer to 16%. (Gusto) Smaller teams, more of them. Solo and very-small-team founding is clearly trending up as well — the share of new startups launched by a single founder has been climbing for years — though the precise year-over-year trajectory is muddier in the public data than headline charts imply, so treat the direction as solid and the exact figures as soft.

The sharpest single data point comes from Y Combinator’s Winter 2025 batch: roughly a quarter of the cohort had codebases that were about 95% AI-generated, letting products that would once have needed six-to-eight engineers ship with a small founding team instead. (ALM Corp) YC’s Garry Tan described the batch growing unusually fast — faster, he said, than anything he’d seen at this stage — without pinning it to a specific weekly number. The institutional appetite tracks the same way: reading YC’s Request for Startups, the partners are openly courting “10-person, $100B” companies — a spec that would have been illegible before this productivity jump, even if you’d treat “codified” as the author’s framing rather than YC’s own words. (Medium / Murray)

The loudest version of the thesis runs further still: a steady drumbeat of AI executives now floats the one-person billion-dollar company and the ten-person, billion-dollar-valuation startup as things that arrive “pretty soon” — outcomes that would be unimaginable without this productivity jump. But that’s a prediction, not a sighting. Hold onto the distinction; it matters later.

Why fewer people per task doesn’t mean fewer people #

The mechanism is the part builders should internalize, because it’s the part that’s counterintuitive. AI here is a labor multiplier, not merely a cost cut. HackerRank’s read is that productivity “expands ambitions” — teams plow the hours they save back into new initiatives rather than just trimming the org chart — and the supply side keeps swelling: GitHub reports a new developer joining roughly every second. (HackerRank)

The historical name for this is the Jevons paradox. Apollo’s Torsten Slok put it plainly in April 2026: “When steam engines made coal more efficient, Britain didn’t burn less coal, it burned more. The same pattern is happening for cheaper legal services.” (Fortune) Cheaper output expands the market for it. And the same piece notes Vanguard’s finding that the 100 occupations most exposed to AI automation have outperformed the rest of the labor market on both job growth and real wages — the opposite of what the “automation eats jobs” story predicts.

At the macro scale, the World Economic Forum’s Future of Jobs work projects something like 92 million roles displaced by 2030 against 170 million created — a net gain on the order of 78 million. (WEF Future of Jobs) Net positive, structurally reshuffled.

The proof is in the cap tables #

Theory is cheap; the empirical anchor is Lovable (Sweden), founded by Anton Osika and Fabian Hedin. The vibe-coding platform launched in December 2024 and hit roughly $100M ARR by mid-2025 — faster than OpenAI, Cursor, or any prior software company on record — then doubled to about $200M ARR by November, on the way to a $330M Series B at a $6.6B valuation in December 2025, with around 8 million users and 25M+ projects built on the platform. (TechCrunch) Separate reporting puts the company at a team of around 15, with the founders’ combined ownership implying roughly $1.6B each. (AI Funding Tracker) Two founders, a team you could fit around one table, a $6.6B valuation. That is the pattern Naval’s thesis says is now structurally repeatable.

The sharper edge case is Base44 (Israel), the kind of solo-founded, profitable app-builder that gets acquired within months on the strength of a handful of people and no outside capital — the small-team output curve bent about as far as it currently goes. And Safe Superintelligence, Ilya Sutskever’s lab, reportedly reached a $32B valuation on a famously tiny org. (TechCrunch) Valuation-per-employee, not just valuation, is the metric that’s breaking.

The honest counter-section #

If you stop here you’ve written an advertisement. The same data carries warnings.

The macro can be net-positive and the distribution can be ugly. Dario Amodei’s “Zeroth World” framing at Davos 2026 is the load-bearing caveat: he sketches a roughly 10-million-person slice of Silicon Valley growing 50% while the broader economy grows 10%, and floats 5–10% GDP growth alongside 10% unemployment — a pairing with no historical precedent. (The AI Corner) “Net 78 million jobs” can be true while the gains pool in a handful of zip codes.

Juniors eat the transition. The “fewer people per task” mechanism has a victim: entry-level workers, the population large teams used to absorb and train. Dallas Fed data cited in the Fortune piece shows employment for workers aged 22–25 in AI-exposed roles down about 13% over the 2022–2025 window, even as overall employment held. (Fortune) Consistent with the thesis, and a direct rebuttal to anyone calling the transition painless.

Trust doesn’t compress. The tiny-team model seems to run cleanest for bottoms-up consumer products that dodge enterprise sales. In B2B, human-to-human trust is still the bottleneck, and building it still costs people-hours AI can’t yet collapse — which is why the lean-team examples cluster on the consumer side rather than in long enterprise sales cycles.

The one-person unicorn is still a forecast. As of mid-2026 no single-founder billion-dollar company has actually landed. The closest cases — the solo-founded app-builders that exit after a few months — top out at valuations two orders of magnitude short of unicorn status. The “very small teams” pattern is real and measurable. The most extreme edge cases are projections wearing the costume of fact.

What to do with this #

The takeaway for a builder isn’t “fire your team.” It’s that the unit of ambition just got cheaper, so the right move is to point the freed-up capacity at the next thing rather than at the exit. The number of viable companies a given amount of talent can spin up went up, not down. The risk that should keep you honest is distributional: a world of many tiny winning teams can coexist with a world that’s stopped hiring the juniors who’d have become the next generation of founders. Both halves of that are in the data. Build for the first; don’t pretend the second isn’t there.