AI Tools Don't Give a Founder the Judgment They Lack

The Myth That Silicon Valley Is Hooked On

A narrative has gained traction in recent weeks: if you can't build your product alone, with AI, you shouldn't be building anything at all. Asking for help, whether by hiring an agency, a contractor, or a first employee, has become a red flag for a growing share of investors. What this narrative hides is a confusion between being able to produce something and knowing whether that something is good.

Dan Maccarone, a columnist for UX Collective, summed up the problem I had seen forming for months: AI has driven the cost of producing software down to nearly zero, but it hasn't driven down the cost of judgment.

We got product creation for free, but forgot that expertise didn't come with it.

Dan Maccarone, author of the original article on UX Collective

This is the position I defend here: AI has democratized execution, not judgment. And it's exactly at this boundary, between discovery and delivery, that the solo founder will stall once the product moves past the MVP (the minimum first version that tests whether the idea has legs).

Where the Pressure to Found Alone Comes From

Paul Graham, founder of Y Combinator, has argued for years that he prefers founders capable of building the product with their own hands. Carles Reina, an early-stage investor at ElevenLabs, made a similar point in a CNBC interview describing his first red flag.

literally cannot build products.

Carles Reina, early-stage investor at ElevenLabs, to CNBC

Today, more than a third of new startups are founded by a single person, according to the Solo Founders report, because AI has made it possible to go from idea to working product without hiring anyone. Dario Amodei, CEO of Anthropic, puts the odds at 70% to 80% that the first billion-dollar company run by a single person will appear as soon as 2026.

He has an obvious interest in this, since Anthropic sells the tool that would make it possible, but the number reflects a consensus that also shows up on the capital side. The Speedrun team at Andreessen Horowitz argues that the question "can you build it alone" no longer matters, because the answer is always yes. The question that remains is "should you build this", which is exactly what Marty Cagan has been repeating for decades about product.

What Startup Genome Had Already Proven Before AI

This appetite for lean teams isn't new or wrong. The Startup Genome Report, analyzing more than 3,200 startups, found that premature scaling, hiring too many people too soon, is the single most common cause of failure.

This logic already worked before generative AI existed. In 1999, a team of eight specialists at the agency iXL took over a forty-person project that had been stuck for months and delivered what the larger team couldn't get off the ground. The idea lines up with Brooks's Law, from 1975: adding people to a late project only makes it later, because every new person pulls whoever already knows how to do the work away to explain what needs to be done. Amazon formalized this logic in the two-pizza team: no team should be bigger than two pizzas can feed.

But there's a huge difference between cutting forty people down to eight specialists and cutting eight specialists down to one founder and a tool. The first move concentrates expertise in a smaller group. The second removes the expertise and hopes the software alone fills the gap left behind. It's worth remembering that most of the giants that didn't make that cut, like MarchFirst, a nine-thousand-person, six-billion-dollar consultancy, burned through a billion and a half dollars in fourteen months and filed the largest bankruptcy in the sector.

When the Solo Founder Really Works

There's one case where building alone is a real advantage, not a gamble. Travis Howell and Todd Hall, in research published in the Strategic Management Journal, showed that solo founders compete on equal footing with co-founded teams under one condition: when the founder already carries deep, broad experience in the craft they're building.

It works when the founder understands the technical risk, the craft detail, what makes a product good, not just the problem it solves. It's the case of the engineer building a tool for other engineers: they are simultaneously the builder and the user, and their knowledge maps exactly onto what the product needs. Outside this specific case, the argument falls apart, because knowing a market or a customer deeply isn't the same as knowing how to build something people actually want to use.

The Blind Spot AI Doesn't Solve

Ethan Mollick, a researcher fairly optimistic about AI at work, insists that it rewards people who already have expertise: only those who know what "good" looks like can catch the confident mistake the model delivers disguised as a finished answer. Tao An, a researcher at Hawaii Pacific University, reached a complementary conclusion. On routine tasks, AI raises beginners to a competent floor, which is real and useful. On tasks that require judgment, it does the opposite and widens the gap between those who know and those who don't.

In short: AI amplifies the expertise that's already in the room. It doesn't supply that expertise.

A concrete example of what this costs: Christian Holst measured that the average e-commerce checkout asks for about eleven fields when eight would be enough, and that nearly one in five people abandons the purchase just because the process feels too long. In testing, cutting the form from eleven fields to four increased completed purchases by 120%. The tool that generated that eleven-field form has no idea it's costing customers. Only someone who has already seen this pattern before knows to ask the question.

The Counterargument I Take Seriously

The pressure for lean teams isn't irrational, and I won't pretend it is. Rebecca Lynn, an investor quoted in the UX Collective article, warns that it has never been easier to put together a company in demo mode, which is why she's more selective about what she funds, not less. Elizabeth Yin and Renata Quintini, at TechCrunch Disrupt, argued that using AI instead of early hiring is the right move to find product-market fit (the moment when the product finally solves a problem the market is willing to pay for) without burning cash on people who aren't needed yet.

I sign off on that part: a small team at the start is a better strategy than a big team too soon, with or without AI. The mistake isn't in starting lean. It's in turning "I started lean" into "I don't need expertise outside my own head", and treating anyone who asks for help as if they had failed before they even began.

What This Changes for Whoever Is Deciding Right Now

If you're a founder putting together your first team, or deciding not to put one together at all, the question worth asking isn't whether you can build it alone with AI. It's whether the expertise that decides if the product is good exists somewhere in the operation, whether that's you, a cofounder, a contractor, or a one-off consultancy. If the answer is no, the MVP will come out looking polished and functional, and it will still stall the moment it meets real people using it, paying for it, and abandoning it.

Paul Rand charged Steve Jobs a hundred thousand dollars for a single logo for NeXT, in 1986, not for the black cube itself, but for the reasoning that explained why that was the one right angle. Jobs could generate a thousand variations on his own. What he lacked, and he knew it, was the accumulated judgment to know which one was the right one. That still separates whoever ships a product from whoever builds something that lasts, except now anyone can get to the first part alone.