This fall, sitting alone at my desk in North Carolina, I started my eighth company with a rule I could not have imagined for the first seven: zero employees. That sounds like a staffing decision. It is not. It is a thesis about what a company is becoming.
Adventure256 will begin with one human: me. There is no human co-founder sitting in the office across the hall and no first wave of employees waiting to be hired. Instead, I am surrounded by models, software and a growing collection of AI agents that can research, analyze, write, code, test, design, organize and increasingly act.
In my early 60s, after decades of building companies through several waves of technology, I find that almost absurd.
It also makes me think about the word founder itself.
We remember Benjamin Franklin, George Washington, Thomas Jefferson, John Adams and Alexander Hamilton as Founding Fathers. Plural. Of course they were plural. Building anything consequential has always required other people because the knowledge, labor and capability required to build it lived inside other people.
For most of history, the idea was the cheap part. Turning it into something real required capital, specialized knowledge, machinery, distribution and usually a collection of people who knew how to do things you did not.
Those barriers have been falling for decades. Now another one is collapsing: capability.
For the first time, a curious person with experience, a computer and access to modern AI systems can begin assembling many of the capabilities of a company before assembling the company itself.
That changes who gets to be a founder.
Founding is separating from hiring
Stripe, the payments and financial infrastructure company used by millions of businesses, published a number in May that made me stop and think. Among C corporations formed through Stripe Atlas so far in the second quarter of 2026, 63 percent had a single founder, an all-time high for Atlas.¹
A month later, Stripe’s economics team found the same pattern at a larger scale: solopreneurship is growing faster than employer-business formation, and early evidence suggests AI is filling capability gaps that once required hiring. In Stripe’s proxy data, more than twice as many solo businesses crossed $1 million in revenue in 2025 as in 2023.²
But the opportunity is not being distributed evenly. Among solo-founded Atlas companies, top-decile founders generated about 34 times the median founder’s revenue during their first six months four years ago. By 2025, the gap had widened to 61 times.
The technology is not making everyone successful. It is making the attempt dramatically more accessible.
AI is lowering the cost of finding out whether you should be a founder.
And something even more interesting is happening underneath those numbers. Founding and hiring are beginning to separate. That distinction did not matter much before. If the business grew, headcount usually grew with it. A founder hit the boundary of personal capability, hired someone, hit another boundary and hired again.
Now there is another option, and it changes something larger than the founder’s to-do list.
The unit that is changing is not the task. It is the organization.
The workshop moved onto the desk
I have seen a version of this pattern before.
In 1985, Aldus PageMaker arrived on the Macintosh. Along with Apple’s LaserWriter and Adobe’s PostScript, it helped create desktop publishing.³ Work that once required specialized equipment and a commercial production process moved onto an ordinary desk.
PageMaker did not make everyone a designer. Plenty of early newsletters proved that. What changed was access to the attempt. You could compose the page, move the type, place the image, print another version, look at what you made and improve it. The cost and friction of experimentation fell.
Years later, I could design a business card while flying at 32,000 feet, send the finished file to a printer from the airplane, and arrange for the physical cards to be waiting at my hotel. The print shop had not disappeared. But a surprising amount of the capability that once lived inside it had moved to me.
Then the web did the same thing to distribution. Cloud computing did it to infrastructure. Software services did it to payments, commerce, communications, analytics and accounting. Each wave moved another institutional capability closer to the individual.
But there was still a limit. I remained the point through which the work had to pass. Better software let me do more, but there was still only one of me, moving serially from one task to the next. AI changes that arithmetic.
The web gave one person reach. AI gives one person an organization.
Parallel changes the arithmetic
This may be the part of the story that is easiest to underestimate because the old language of “productivity” does not quite describe it.
I work in parallel now. While I am thinking through one problem, an agent can be researching another. Another can be building software. Another can be testing what was just built. Another can compare approaches or review the work of the first three.
This is becoming a recognizable operating model. The industry language is multi-agent orchestration, concurrent workflows and parallel agents. OpenAI now explicitly describes multi-agent systems as a way to split independent work into separate streams that proceed at the same time. Anthropic has published experiments using teams of agents working concurrently on a shared technical objective.⁴,⁵
The internal usage numbers at OpenAI make the point more vividly. By June 2026, the heaviest Codex users were generating more than 60 hours of agent work in a single day, distributed across multiple parallel agents.⁶ By mid-August, OpenAI’s research organization was using 3.1 agent-workdays for every human workday.⁷
More machine workdays than human workdays.
That sentence would have sounded nonsensical not long ago. Now it describes an emerging way of working. Humans still decide what matters. We set priorities, judge quality, recognize when the answer misses the point, take responsibility and decide what happens next. But work no longer has to queue behind one pair of hands.
That is why “AI makes people more productive” feels increasingly inadequate to me. It begins to look like organizational capacity. And it leads to a question I cannot stop thinking about:
How large can one human become?
Experience can now become a company
Imagine a hotel general manager who has spent twenty years watching the same operational failures repeat.
A guest tells the front desk the air conditioner is making a terrible noise. Someone texts maintenance. The shift changes. Housekeeping notices the same problem the next morning. A manager apologizes to the guest, but the property now has four slightly different versions of what happened living in conversations, texts and systems. The general manager understands the problem because she has lived it hundreds of times.
Five years ago, that understanding was not enough to build a company around it. She might have needed a technical co-founder, software developers, capital, a product team or a vendor willing to care about one irritating operational problem.
Now she can start differently. She can use AI to research the products already in the market, interview the problem from multiple angles, map the workflow, build a rough prototype, load it with sample information and put it in front of three other hotel operators.
One tells her it is useless. Good information. Another ignores the feature she thought was brilliant and points to something she almost left out. The third asks what it costs. Now she is learning from reality.
That example could be almost anyone with accumulated experience:
A nurse who knows exactly where patient information gets lost.
A property manager who has solved the same tenant problem five hundred times.
An accountant who knows why a certain type of small business closes its books badly every month.
A teacher who discovered a better way to explain the thing students never understand.
A hotel operator who knows what guests ask for at 11:30 every night.
They already possess the scarce ingredient: they know something. We have spent enormous energy asking whether AI will make human experience obsolete. I think we may be looking through the wrong end of the telescope.
Experience can now become a company.
That is the opportunity because AI can learn a programming language. It can research a market. It can generate pricing models, build software and propose ten approaches before breakfast. But it did not spend twenty years noticing the problem. You did.
I have learned most of what I know the same way: by experimenting. Make something. Put it in the world. Discover what you misunderstood. Change it. Try again.
You learn by making the thing.
Reid Hoffman, the co-founder of LinkedIn, captured the startup version of this years ago: “If you’re not embarrassed by the first version of your product, you’ve launched too late.”⁸
That does not mean ship garbage. Hoffman has been explicit about that. It means do not confuse polishing with learning. Ship early enough to learn.
The experiment
I recently finished a chapter as a Chief Innovation Officer. What I am building now is partly a business and partly an experiment in the same forces I am advising other leaders to understand.
How much work can I responsibly delegate to agents? Which kinds of work improve when several agents attack the problem at once? Where does a human specialist make the result materially better? Where am I still acting as the bottleneck simply because I have not learned to hand the work away?
And what remains stubbornly human? Trust is one answer.
An agent can research a customer. It can prepare a proposal, analyze a meeting, write software and monitor a system. But a customer still decides whether to trust me. I still decide what I am willing to put my name on. Responsibility does not disappear because the work was delegated to software.
Capability is becoming abundant. Trust is still human.
I do not have all the answers to the one-person-company experiment. That is the point. I have lived through enough technology cycles to know that clarity always arrives after you begin, not before. The work teaches the experiment. The experiment teaches the work.
And because the cost of testing an idea has fallen so far, an experiment can stay small without thinking small.
Small experiments can now have large ambitions.
Your new co-founder
I think almost everyone should consider becoming a founder.
That is deliberately provocative, but I do not mean everyone should quit a job, raise venture capital or build another app looking for a problem. I mean founder in the older and more useful sense of the word: find something worth creating and begin.
For many businesses, the opening bet can now be remarkably small: a computer, an internet connection, a handful of subscriptions, some useful experience and the curiosity to keep asking better questions. You may not need a technical co-founder just to discover whether the idea works.
AI cannot legally own shares in your company. It cannot sign the operating agreement. It cannot accept moral responsibility for a decision or look a customer in the eye and mean what it says. But functionally, it can increasingly occupy the second chair.
It can help explore the idea, research the market, develop the strategy, create the first product, build the website, prepare sales material, organize operations, analyze feedback, manage systems and coordinate other agents.
Agents are digital workers.
Put enough of those capabilities together and they stop looking like another collection of software tools. They begin to look like organizational capacity. That was not available to an ordinary individual at this scale before. It is now.
Founders, plural
This brings me back to Franklin.
In 1731, long before he helped found a country, Benjamin Franklin and members of his Junto (a Philadelphia club devoted to shared ideas and self-improvement) created the Library Company of Philadelphia.
Books were expensive and difficult to obtain. Many had to cross the Atlantic. No individual in the group could afford the collection they wanted, so they pooled their resources.
Forty shillings to join, then ten shillings a year.⁹ In very rough 2026 purchasing-power terms, think roughly $400 to join and about $100 a year.¹⁰ The important part is not the exact conversion. Franklin had found a way for an individual to reach beyond what he personally owned.
Nearly three centuries later, I look at a credit card statement full of software subscriptions and see a strange echo of the same idea. Only now the shared resource is not a shelf of books. It is capability: research, coding, analysis, design, strategy and increasingly action are available on demand.
Franklin later became one of the people we call the Founding Fathers, along with Washington, Jefferson, Adams, Hamilton and others. Founders, plural. They needed one another. Every consequential venture did. Knowledge lived in people. Skills lived in people. Labor lived in people.
That is still true, and humans will remain essential wherever judgment, trust, accountability, empathy, taste and real expertise matter. But it is no longer the whole truth.
Now one person can sit down with an idea and surround that idea with extraordinary amounts of capability before hiring the first employee.
It does not eliminate people. It changes what one human can begin.
Start something
So when I say everyone should be a founder, I am not prescribing a career. I am arguing for an experiment.
Somewhere in your accumulated experience is probably a problem you understand better than most people. Something inefficient. Something expensive. Something irritating. Something everybody around you has learned to tolerate because fixing it once required resources you did not have.
Maybe you have said the sentence yourself: Someone should build something for that.
The economics of that sentence have changed.
The first experiment may cost less than a nice dinner. The first prototype may take a weekend instead of a funding round. The first collaborator may be an AI system rather than a person you have to convince to quit a job. Most experiments will not become companies. That is fine. The failure is cheap and the learning is real.
But a few will become something useful. A few will find customers. Some will grow. And some person reading this may discover that twenty years of experience was not merely a résumé. It was raw material.
Alan Kay, one of the pioneers of personal computing, said it beautifully: “The best way to predict the future is to invent it.”¹¹
There has never been a better time to take that literally.
Start something!
-DS
Sources & verification notes
Stripe Atlas, “Solo founding is at an all-time high,” May 28, 2026
Stripe Economics, “The age of the solopreneur,” June 22, 2026
Anthropic, “Building a C compiler with a team of parallel Claudes,” Feb. 5, 2026
OpenAI, “Research acceleration: The view inside OpenAI,” Sept. 2026
Library Company of Philadelphia, history of its 1731 subscription model
The $400 / $100 figures are approximate 2026 purchasing-power comparisons, not precise equivalents for colonial Pennsylvania currency. See University of Chicago Press on Pennsylvania currency and long-run U.S. purchasing-power data. Different historical-value methods produce different answers.
Great read, Deven. While AI is scary to some, I love the statement “Responsibility does not disappear because the work was delegated to software.”