The Minimum Viable Firm Has Been Collapsing

I’ve built working software without knowing how to code.

Not a company, and nothing I sell. Tools I built for myself, with AI doing the parts that used to need engineers I couldn’t have hired and wouldn’t have known how to direct. Three years ago, that was a fantasy. Look at how fast it stopped being one.

So, when the The Wall Street Journal recently profiled elite students skipping Wall Street and consulting to spend their summers in San Francisco hacker houses chasing AI startups, I didn’t read it as a story about ambitious kids. I read it as a story about math. The old risk map was simple. A corporate job was safe. A startup was risky. Going solo was fringe. To my excitement that map is breaking as anticipated. Now that said – I don’t think everyone should drop out and go start something, and most who try will fail (done that a few times). But the reason it’s breaking is simple: the cost of starting a company collapsed.

Here’s what I think is happening: AI is shrinking the minimum viable firm.

A company that used to need engineers, marketers, analysts, salespeople, and support staff can now be started by one person or a very small team. Not always well, and not always defensibly. But far more often than before. This isn’t a productivity story. It’s a story about the firm itself.

Why companies existed in the first place

Firms exist because capability was distributed. One person could code but not sell. Sell but not price. Price but not run support. So, you hired people. Then you hired managers to coordinate the people. Then you bought software to coordinate the managers. That was the model, and it held up because the alternative wasn’t truly an option.

AI changes the alternative. One person can now rent capability by the piece: research, drafting, code, design, analysis, support, pricing, competitive intel. The judgment still must be yours. So does the read on the customer. But the cost of getting started dropped, and when that cost drops, behavior changes.

And the data isn’t anecdotal anymore.

Stripe’s economics team says solopreneurship is now growing faster than employer-business formation, and that recent cohorts hit revenue milestones sooner than older ones. The share of businesses reaching $1 million in cumulative revenue within a year ran about 30% higher for the 2025 cohort than for 2023, and roughly three times higher than for 2019. Stripe’s solopreneur proxy index found more than twice as many solo operators cleared $1 million in 2025 as in 2023. Stripe Atlas adds that solo founders made up 63% of the C corps formed through Atlas so far this quarter, an all-time high.

Now consider the source. Stripe makes money when more businesses form and more payments move.

The academic piece is the one most should read. A June 2026 working paper from Hyunjin Kim at INSEAD and Rembrand Koning at Harvard Business School finds that AI-native firms run 25% smaller than comparable startups in the same cohort, carry more engineers, employ fewer entry-level workers and fewer managers, sit about half a seniority level flatter, and still hold comparable valuations. Smaller, flatter, more technical, worth the same. That’s a different kind of company, not a trimmed-down copy of the old one.

And let’s stay honest about the pace. Census data shows only about 17% to 20% of U.S. businesses were using AI through the middle of 2026, with the big firms and the knowledge-heavy sectors out front. Nobody woke up as an AI-native company overnight. But platform shifts get decided at the frontier long before the average catches up.

So now the platform data, the public business-formation numbers, and the academic work all point the same direction, calling it hype is incorrect.

The real mechanism is coordination

Most people still talk about AI as a faster tool for the same work. That misses it and frankly drives me bananas. The power isn’t speed. It’s that AI compresses coordination, and coordination is most of what a firm does.

In The Invisible Interface I call this the move from operating software to delegating outcomes. The old way: people learn tools, do the glue work, and push information across rigid workflows. The new way: a person states intent, and a layer between intention and execution turns it into action across systems and hands back a result you can check. Microsoft put a number on the tax. Its 2025 research found that the most-pinged workers get interrupted every two minutes during core hours, 275 times a day, by meetings, emails, and chats. Take that out and the shape of the work changes, not just the speed.

This is the part the jobs debate keeps getting wrong IMHO. AI isn’t only competing with employees. It’s competing with the firm itself, with the whole idea of gathering people together to get something done. That’s the better way to read the labor story, too. The first shock may not be mass unemployment. It may be migration: people walking out of firms because one of them can now do what used to take a small team.

That’s why the solopreneur matters even if you’ll never be one. Solo operators are the far edge of the same shift, running with no bureaucracy to unlearn. They’re the lab. The rest of us will run the same experiment inside our own walls, slower, and against a lot more resistance/friction.

And this is where most enterprise AI programs are kidding themselves. Buying licenses isn’t transformation. Running pilots isn’t deployment. Telling people to use AI more isn’t a strategy. Hand your team tools and you’ve made the old firm a little faster. Move the capability into the product and the workflow, and you’ve started to rebuild the firm. Kim and Koning name both roads: a process channel that changes how people work, and a product channel that puts the capability into what you sell. The second one is what changes your company.

Building got easier. Getting chosen got harder.

Here’s the part the solopreneur cheerleaders may skip. The same AI that lets one person build a company can also make that company invisible.

In the old digital economy, people browsed, compared, and clicked. In a market where agents do the choosing, the customer may never browse. They tell an agent to find the option, weigh the tradeoffs, and buy the one that fits. Your website turns into a data source. Your brand turns into a machine-readable claim. Your funnel turns into an API call. When an agent shows three options, options four through forty didn’t lose. They were never seen. I tend to ask CEOs I know – if you end up there – do you exist?

So, the advantage moves to whoever the agent can read, verify, and safely buy from. Unclear data is the new slow website: it kills the sale before you even know you were in the running.

And independence isn’t what it looks like. The one-person company runs on rented cognition, rented payments, rented infrastructure, rented distribution, rented memory. Platforms are leverage, and I lean on them too. But every rail can turn into a toll road. In the book I call this cognitive rent: the tax you pay because someone else holds the memory you depend on. Your preferences, your patterns, your relationships, your decision history. Whoever holds that context controls your switching costs, your defaults, and eventually your price. If you can’t leave without losing the thing that makes you productive, you don’t own it. You’re renting it, and the landlord sets the rent.

What I’d do if I was running a large company right now

Not a solopreneur. A hundred-person or hundred-thousand-person company watching a three-person competitor go straight for the part of the business where customers pay for friction you’ve quietly accepted as normal.

Run the threat map. Where could three people with AI deliver 80% of what you deliver at 20% of the cost? Answer with the specific workflow, not with ‘our brand’ or ‘our relationships.’

Audit the coordination tax. Find the work where people are moving information around instead of using judgment. That’s where AI can help first.

Stop buying tools and start killing workflows. The question isn’t which AI tool to buy. It’s which workflow shouldn’t exist in its current form.

Protect the apprenticeship. Face the fact that if AI eats the junior work that used to train your people, you must rebuild that learning some other way. Lean isn’t the same as resilient. A company can get faster and more fragile at the same time.

The firm was built to coordinate people, capital, information, trust, and risk. AI changes the cost of coordination, so the reason the firm exists changes with it. Firms don’t disappear. But the winners won’t just be the smallest companies. They’ll be the ones, at any size, that move as fast as a solopreneur without getting as fragile as one.

The minimum viable firm has been collapsing. Now you will find out which kind of organization you’ve built: smaller, smarter, and more capable, or a layer of human middleware waiting to get routed around.

Sources

• Stripe Economics, “The Age of the Solopreneur” (June 2026)

• Stripe Atlas, “Solo founding is at an all-time high” (May 2026)

• Kim (INSEAD) & Koning (HBS), “AI-Native Firms” working paper (June 2026)

• U.S. Census Bureau, AI use at U.S. businesses (BTOS, 2026)

• Microsoft, “Breaking Down the Infinite Workday,” Work Trend Index special report (June 2025)

• Wall Street Journal, elite students and AI startups

• Wall Street Journal Opinion, Liya Palagashvili, “Me, Myself and AI”