The Stunt and the Story Behind It
You've probably seen the headlines. A founder builds a software company staffed entirely by AI agents. Another launches a product studio with zero full-time employees and a revenue number that looks impressive until you check the margin structure. FelixCraft reported over $2 million in annual recurring revenue with two humans and a stack of autonomous agents. Pulia ran a content production operation where a single human monitored outputs and handled escalations while agents did everything else.
These are genuinely interesting experiments. And they make for extraordinary content. But calling them a business model is like calling a Formula 1 car a commuter vehicle. Technically, it moves people from one place to another. Practically, it's not built for the conditions most of us operate in.
Why the Zero-Employee Model Doesn't Scale the Way It Looks
FelixCraft's numbers are real, but the product is narrowly scoped, the customer base is technically sophisticated, and the edge cases that break agent workflows simply don't appear as often in that context as they would in a professional services firm or a specialty brokerage or a logistics company with complex supplier relationships.
Pulia is more candid about this. Their team has been public about the volume of edge cases requiring human intervention — not because the agents were bad, but because the real world generates exceptions at a rate that no agent stack, as of early 2026, handles gracefully without human oversight. The value wasn't in removing humans. The value was in understanding exactly where human judgment was irreplaceable and building the system around that insight.
Small Teams Punching Way Above Their Weight
While the zero-employee narrative captures attention, something quieter and more durable is happening. Teams of eight to twelve people are routinely doing work that required thirty to fifty people five years ago.
McKinsey's research on generative AI adoption found productivity gains of 20 to 25 percent on average across organisations past early experimentation. But the distribution is not normal. Teams with intentional AI workflow design — where AI handles the repetitive and complicated, and humans focus on judgment, relationships, and exceptions — are seeing gains well beyond that range.
In practice, AI handles the work that benefits from speed and volume: pulling competitive intelligence, drafting documents from a brief, summarising meetings, processing inbound inquiries. Humans handle the work requiring context, judgment, and trust: deciding whether intelligence changes the strategy, knowing which client needs a phone call instead of an email, reading the room in a negotiation.
The best-performing small teams have built their AI layer to absorb 60 to 70 percent of the volume while humans own 100 percent of the consequential decisions.
How to Build Toward It
Map your existing workflows against two questions: where is the volume concentrated, and where is the judgment irreplaceable? The first category is your AI opportunity. The second is where you protect and develop your humans further.
Most organisations find that two or three workflows contain the majority of the volume and the least irreplaceable judgment. Start there. Build something that works for a month before you expand. Document the edge cases you actually encounter — they're almost never the ones you anticipated.
The teams figuring this out aren't chasing the most impressive AI stack. They're designing the division of labour clearly, building human capability around the work that matters most, and letting AI take the volume so humans can take the value. That's not a stunt. That's a business.