Only One Has No Employees

I run three companies. At one, I am the only person — no employees, no co-founder, nobody. At the other two, we are three. For a year I treated the solo venture as the disciplined choice and the two team ventures as the normal ones. That distinction turned out to matter less than I assumed. What actually makes all three possible isn't the headcount. It's the same system running underneath, regardless of how many people are technically on the org chart.

The solo operator is no longer the exception

The U.S. Census Bureau counted roughly 30.4 million nonemployer businesses in 2023 — enterprises with no paid staff — with gross receipts of nearly $1.8 trillion that year, equivalent to about 6.4% of U.S. GDP. [1] Between 2012 and 2023, those businesses grew an average of 2.7% a year, more than double the 1.1% growth rate of businesses that do hire. Nonemployer businesses have been the majority of U.S. establishments for decades, and that gap keeps widening — a company with no org chart at all is the normal case, not the edge case, and my short-term rental operation across Portugal and Spain is exactly that: zero employees, by design and by necessity.

The advice written for that person is still wrong

Almost everything published for solo operators treats AI as a productivity multiplier — write the email faster, automate the invoice, save two hours on a Tuesday. That framing carries a hidden assumption: that the business would exist and run without it, only more slowly. For the property management company, that assumption never held. Two legal systems, two tax regimes, guest communication in five languages, compliance obligations that shift depending on which side of a regional border a listing sits on — none of that shrinks because one person is tired at 11pm.

At my other two ventures — a product and AI consultancy, and a healthtech company navigating EU medical device regulation — the gap shows up differently. The three of us each have real specialisations; this isn't a generalist team papering over missing roles. But specialisation was never the same problem as capacity. Three people, however sharply divided by expertise, still don't have the hours of a larger org, and still don't have a built-in second reviewer sitting outside their own specialism for the work that falls between lanes. "We have a team" answers the expertise question. It doesn't answer the bandwidth one.

What actually runs underneath

What sits under all three ventures now is not a set of shortcuts. It is closer to a standing team that never sleeps and never forgets what it decided last week: an agent system that drafts and checks guest messages in five languages before I ever see them, cross-references a compliance question against the actual regulation text rather than my memory of it, builds and stress-tests financial models that would once have required an accountant or analyst none of the three ventures could yet justify hiring, and keeps a persistent record of every decision so the next question doesn't start from zero.

Take one concrete example from the venture I run alone. Portugal introduced a new capital gains reinvestment relief on property sales this year — a gain can be exempt from tax if the proceeds go into another residential property, provided that property is then placed under a qualifying residential lease of at least 36 months (consecutive or with interruptions, within a five-year window), at a rent capped relative to the 2026 minimum wage. The decree does not address short-term rentals by name. But a property operated exclusively as short-term accommodation, with no qualifying residential lease in place, would not meet that separate condition — a distinction only visible from reading the decree text itself, not from the headline coverage. Working it out alone, at midnight, from memory of a headline half-remembered, would have produced the wrong answer with total confidence. Instead it ran through a structured read of the actual decree text, cross-referenced against the relevant regulatory detail, logged so the reasoning is retrievable the next time the topic comes up rather than re-derived from scratch.

None of this replaced a hire in the way people usually mean it. At the solo venture, there was no hire it was replacing — a compliance associate, a bilingual guest-relations coordinator, a junior financial analyst were never going to exist in a one-person structure at this stage, at any price I could pay. At the two team ventures, it replaced the fourth, fifth, and sixth person three of us would otherwise have needed before the company could operate at this level. Different shape, same function: the choice was never "AI versus people." It was "AI versus those functions simply not existing yet."

The part I didn't expect

I expected the AI to save time. What I didn't expect is that the highest-stakes work — the compliance cross-checks, the financial models, the legal reads — got more rigorous, not less, once a second, independent pass entered the loop, at all three ventures alike. One person reviewing their own work at midnight has a single set of eyes and a strong incentive to see what they want to see. Three people under deadline pressure aren't much better — everyone assumes someone else already checked the thing they didn't have time to check. A system with no stake in being right, only in being accurate, catches what either version of "not enough people" misses.

I have caught real numbers this way that no one on any of the three teams would have caught alone. A statistic drafted for a client-facing article was quietly cut before publication because a second, independent check found no reliable source behind it — the number simply did not survive scrutiny it would never have received from a single tired read-through. That is the pattern, repeated across compliance, finance, and content — a review structurally independent from whoever wrote the first draft, which neither a solo operator nor a three-person team can otherwise arrange without adding headcount. The risk I spent a year worrying about — the AI getting something wrong — turned out smaller than the risk I hadn't named: too few people, at any of the three, ever getting a genuine second opinion.

The part that matters going forward

For the venture I run alone, AI is not optimizing an existing operation at the margin. It is the operating capacity itself — remove it, and that company does not get slower, it stops existing in its current form. For the two ventures where we are three, it is not a replacement for expertise — we have that — but a multiplier on capacity, letting three specialists cover the bandwidth and cross-checking a much larger team would otherwise need to provide. Either way, the number of names on the org chart stopped being an honest description of who — or what — is actually doing the work.

At what point does a tool stop being a tool and start being the colleague nobody put on the org chart?

Sources

1. U.S. Census Bureau, Number of U.S. Nonemployers Grew Faster Than Employer Businesses Nearly Every Year From 2012 to 2023, based on the 2023 Nonemployer Statistics release — census.gov, May–July 2025. https://www.census.gov/library/stories/2025/07/nonemployer-business-growth.html
2. Diário da República, 1.ª série, n.º 97 — Decreto-Lei n.º 97/2026, de 20 de maio (capital gains reinvestment relief, article 5.º introducing article 10.º n.º 7-9 of the Código do IRS). https://diariodarepublica.pt/dr/detalhe/decreto-lei/97-2026-1124493227
3. Personal operating record — LusiberiaStays (property management, Portugal/Spain, solo), THE iCON (AI product and consultancy, 3-person team), Q-LIRI (healthtech, EU MDR, 3-person team), 2026.