Half a House
A one-person business isn’t small. It’s missing four departments.
10 Nov 2026 · Manifesto · Andrei · Founder @ Jinna.ai
Thirty million Americans run a business alone. My bet is that Gen Z’s hunger for freedom, given wings by AI, will double that number over the next ten years. The default company of the next decade has one person. But almost all business software is still built for the other kind.
I bought my first computer at 20. It cost half as much as a house in the place where I grew up.
I was born in the Soviet Union and lived 30 km from Chernobyl when the nuclear disaster happened. I grew up in a village, feeding pigs, cutting hay, and herding cows until 18. A house was the biggest purchase anyone there would ever make. My first money came from selling 300 Stephen King books in one day, door to door across a university campus.
Then I handed half a house to a shop for a beige box, because I believed the small box could do what the big one never would.
It could set me free.
Freedom has always been practical for me. The ability to leave. To choose what to work on. To try an idea without waiting for someone to hire me to try it. Since that computer I’ve moved between countries, built four businesses and started over more than once. Before any of that, at my first job I sold advertising to local businesses from a phone booth, because our apartment had no phone. My first company sold websites and advertising to the smallest companies in town. A hundred and twenty of them signed up in the first year.
Later I spent years inside fast-growing companies, Revolut, Bolt and Osome, with big teams and real budgets. At two of them I built finance and admin tools for small businesses. They were good tools, and the owners still did the work after hours. Now I’ve come back to the smallest companies on earth. The clearest view I have of them is a coffee shop.
The second job
My wife, Alya, runs a café. It’s where the neighborhood meets before work. It’s small enough that she is behind the counter herself most days. One evening last week she locked up at five, after ten hours of that, and went home to the rest of it: ordering supplies, getting papers ready for the accountant, emailing the landlord about the lease.
An investor would call it a lifestyle business. I’d call it the economy. Cafés, florists, the person who walks your dog, the small gallery, the architect redoing the house down the street. They give a street its character, and their owners keep very little of the time or money for themselves.
The coffee is the part Alya signed up for. The rest is a second job nobody applied for: next week’s schedule, the supplier who didn’t deliver, payroll, the laptop after the shift. She didn’t open a café to do that job. She opened it to be free. The second job is where the freedom went.
She was right
Not long ago I pitched Jinna, the back office I’m building for one-person businesses, to an investor. I hadn’t finished explaining what it was when she stopped me. “Everyone knows this segment is unattractive,” she said. Then she listed why. Small checks. High churn. People who sign up, use a thing twice and disappear.
She wasn’t being unkind. She was reciting the consensus, and the consensus is correct. I’ve sold to these businesses and built software for them for most of my working life, and I’ve seen every item on her list.
Her list wasn’t even complete. A customer with a small budget can still need a lot of support, a business can close for reasons no product can fix, and selling to thousands of individual owners costs money. Caring about these people doesn’t make any of that disappear. Small business has been a bad software market.
It was never a software market
It was never a software market
A one-person business has almost no software budget. It pays for labor instead, an accountant and maybe an assistant for a few hours a week. Whatever is left, the owner does herself in the evening. Alya pays her accountant about $235 a month. All her software together costs less than $50. The rest she pays for in evenings.
In the United States alone, about 30 million businesses have no employees. That’s close to four out of five of all businesses in the country. The number usually grows 2–3% a year. In the latest count it grew 5%. For thirty years the software industry looked at this market, saw the subscription line, which is small, and missed the labor line, meaning the money owners pay other people and the hours they put in themselves.
Software could only sell these people tools, and a tool leaves its owner with a job. She has to learn the interface, enter the information, decide what happens next, check the result and come back when something changes. Put five useful tools together and the owner gets a sixth job: keeping them consistent. In a larger company that work is shared, but in a one-person business it comes back to the same person.
Software sold to a solo business also costs the owner’s attention. Small checks are a fair price for something that saves 20 minutes and costs attention to run. Churn comes from the same place, because you abandon a tool the day you stop feeding it.
AI changes what can be sold. A system that knows the business, remembers what was agreed and finishes the task is a share of a job. And a share of a job is worth what the owner used to pay for it, in money or in evenings. That is a different number from the price of a subscription.
It’s a bet, and it has a ceiling. The average business in that count brings in roughly $60,000 a year, so nobody here will pay a salary for software. My bet starts with the owners who quote before they work, because a won proposal pays for the help that wrote it. Today Jinna is priced like software, by the month, and that is the right price for now: the work hasn’t yet earned a labor price. Willingness to pay more has to be demonstrated, one finished job at a time. But I’m confident of the frame. The investor was right that small business is a terrible software market. It was always a labor market.
Four departments missing
Every business, whatever it makes, runs the same loop: find the client, agree on the work, do it, bill it, get paid. The day I sold those 300 books, I ran all of it on foot. A large company hires a person for each step. A one-person business hires nobody and does all of them, after hours, and nobody is good at five jobs at once.
We call that a small business. It isn’t. It’s a full-sized company with four departments missing, staffed at night by one tired person.
Thirty million of these companies already exist. What’s new is that one person’s business can now be as complete as a staffed one. The agreement is kept, the invoice is right and the payment is chased, without the owner carrying every step in her head. One person still decides everything, and the rest is carried.
Companies grew, in large part, because functions needed heads. If functions stop needing heads, the natural size of a business becomes whatever its owner wants it to be. Some owners will want to grow. Many will want to stay one person, by choice, with help. Alya is happy behind the counter. She never wanted the admin. She wanted it gone.
The human was the integration layer
Why did single-purpose tools win for so long? Because context was free. The owner carried it. She remembered that this client agreed to $1,200 with work starting after the fifteenth, that he pays late, that the last invoice had a discount, and she kept all of it in her head, between the proposal app, the invoice app, the to-do list and email. Every specialist tool was built in isolation and assumed a human would glue it to the rest.
Consider a photographer who has agreed to shoot a wedding. There’s a date, a price, a deposit, a scope and a balance. The couple asks for an extra hour, and the photographer says yes. Now several things have to stay consistent. An AI can write a lovely reply and leave the old price in the proposal. It can issue the right invoice and then send a reminder after the money has arrived. Each output looks plausible while the job, as a whole, goes wrong.
An agent hasn’t taken work off your hands if you still have to keep it in your head.
The moment the owner stops carrying context, the old arrangement breaks. The context has to live somewhere, and wherever it lives becomes the real product. In Jinna it lives in a record of what was agreed. Once the photographer updates the proposal with the extra hour, the invoice for that wedding takes its amount from it, never from what the model happens to recall.
A proposal written by a system that remembers the last invoice beats one written by the best proposal tool on earth that doesn’t. That is my argument against five separate apps. There will still be excellent specialist tools, and an agent may use several. The boundary that matters is who carries the context between them, and putting five apps behind one login doesn’t move it. An all-in-one suite files the documents. The agreement keeps moving after they’re signed: an extra hour, a later date, a discount given on the phone. Someone has to hold the current version, and today that is usually the owner.
What it takes to carry a job
Cheap model calls can produce expensive supervision. If the owner has to check, repeat and repair, that time belongs in the cost of the work, and a system that needs constant watching hasn’t changed the economics at all. So everything above depends on reliability, and reliability turns out to be mostly about memory, permission and checking.
A business has facts, and it has promises. We agreed this price. We said Friday. This client has an exception. This invoice waits because the scope isn’t approved. Memory has to keep the difference between a passing suggestion and an agreement, between the usual policy and a specific exception, and it has to know which version is current. Otherwise remembering more just makes the system confidently wrong in more ways.
Then the memory has to reach the action. This one I learned the hard way. For a while Jinna could save a business’s rules, and the part that wrote documents never saw them. The memory lived in the chat, and a proposal could be drafted that ignored every rule the owner had just confirmed. Remembering is cheap. Remembering at the exact moment a document is written is the product.
It doesn’t take much. What Jinna puts in front of the model about a business is a card of roughly five hundred tokens, a few hundred words, a small fraction of what the model reads each time, and the work is in choosing them.
Permission has to be specific. Drafting a proposal, sending a routine reminder, changing a price and committing someone’s Saturday carry different consequences. Being right about the client’s name doesn’t earn the right to offer him a discount. We built Jinna never to save its own guess about a business without asking, and to show the owner everything it knows. Autonomy has to be earned one fact at a time, and the owner must always be able to see what happened and why, and change the instructions.
The system also has to check its own work before the owner sees it. Jinna re-reads the saved document before it quotes a figure from it. When a client’s name matches two invoices, it asks which one.
One more condition has nothing to do with reliability. The memory has to be portable. If the accumulated understanding of your own business is something you can’t inspect, correct or take with you, your independence has acquired a new dependency. In Jinna the owner can already see and correct all of it. Taking it with her is the part we still owe. What keeps a customer should be that the system knows the business, not that she can’t leave.
What we're building
Jinna is the brain and the hands of a one-person business. The brain is a memory of your business that you can see and correct: what you’ve agreed, what you charge, how you work, who owes you what. The hands are the parts of the loop it carries. Today that means proposals, estimates, invoices, the emails around them and the reminder when a payment is late. They are learned from the owner and from the business’s own website, written in the owner’s terms and sent when the owner says so. That isn’t Alya’s café yet. It’s the photographer, the architect, the consultant: anyone who quotes before they work.
That’s the starting point. More than 2,500 people have signed up, and nearly $5 million in proposals, estimates and invoices has run through Jinna. The version built around memory shipped this month. The end state is a system that runs the whole loop, from a client’s first contact to money in the bank, within whatever authority the owner chooses to hand over. I think that’s years away, not months.
If the whole loop works, Jinna is the back office for millions of companies in one country alone. The freedom to stay small belongs to their owners. The company that serves them won’t be small.
None of this is a free pass on the business model. We still have to reach customers at a sensible cost and earn weekly use. Taking on more of the work raises the price of getting it wrong. The questions I want answered by building and measuring are plain. What share of a job gets completed reliably within the owner’s instructions? How much attention do the exceptions consume? Do people come back in the ordinary week after the demo? The demo is the beginning. The ordinary week is the test.
Two things I don’t know. The first is how fast the models close the gap between “usually right” and “trusted with money”. My guess is sooner than the skeptics think and later than the demos suggest. The second is which parts of the second job stay human. My current answer is the promises. A system can draft the proposal. Only the owner can decide what she is willing to promise, and to whom.
Half a house, again
The computer I bought at 20 cost half a house, and it did almost nothing on its own. Everything it made, I made through it, after hours, the way owners of one-person businesses have worked ever since.
Today the raw capability costs about what one regular spends at Alya’s counter in a month. The system that turns it into a colleague, one that remembers, acts within its permissions and finishes, is what still has to be built. That’s the gap I’m working in.
People who start a business are betting on their freedom. For most of history the bet came with a second job attached, the one that ate the freedom it was supposed to buy. It won’t have to for much longer. People should be able to build something of their own and still have a life of their own.
I’m building for the day Alya locks up, goes home, and there’s nothing left to do.
Half a House
A one-person business isn’t small.
It’s missing four departments.
10 Nov 2026 · Manifesto
Andrei · Founder @ Jinna.ai
Thirty million Americans run a business alone. My bet is that Gen Z’s hunger for freedom, given wings by AI, will double that number over the next ten years. The default company of the next decade has one person. But almost all business software is still built for the other kind.
I bought my first computer at 20. It cost half as much as a house in the place where I grew up.
I was born in the Soviet Union and lived 30 km from Chernobyl when the nuclear disaster happened. I grew up in a village, feeding pigs, cutting hay, and herding cows until 18. A house was the biggest purchase anyone there would ever make. My first money came from selling 300 Stephen King books in one day, door to door across a university campus.
Then I handed half a house to a shop for a beige box, because I believed the small box could do what the big one never would.
It could set me free.
Freedom has always been practical for me. The ability to leave. To choose what to work on. To try an idea without waiting for someone to hire me to try it. Since that computer I’ve moved between countries, built four businesses and started over more than once. Before any of that, at my first job I sold advertising to local businesses from a phone booth, because our apartment had no phone. My first company sold websites and advertising to the smallest companies in town. A hundred and twenty of them signed up in the first year.
Later I spent years inside fast-growing companies, Revolut, Bolt and Osome, with big teams and real budgets. At two of them I built finance and admin tools for small businesses. They were good tools, and the owners still did the work after hours. Now I’ve come back to the smallest companies on earth. The clearest view I have of them is a coffee shop.
The second job
My wife, Alya, runs a café. It’s where the neighborhood meets before work. It’s small enough that she is behind the counter herself most days. One evening last week she locked up at five, after ten hours of that, and went home to the rest of it: ordering supplies, getting papers ready for the accountant, emailing the landlord about the lease.
An investor would call it a lifestyle business. I’d call it the economy. Cafés, florists, the person who walks your dog, the small gallery, the architect redoing the house down the street. They give a street its character, and their owners keep very little of the time or money for themselves.
The coffee is the part Alya signed up for. The rest is a second job nobody applied for: next week’s schedule, the supplier who didn’t deliver, payroll, the laptop after the shift. She didn’t open a café to do that job. She opened it to be free. The second job is where the freedom went.
She was right
Not long ago I pitched Jinna, the back office I’m building for one-person businesses, to an investor. I hadn’t finished explaining what it was when she stopped me. “Everyone knows this segment is unattractive,” she said. Then she listed why. Small checks. High churn. People who sign up, use a thing twice and disappear.
She wasn’t being unkind. She was reciting the consensus, and the consensus is correct. I’ve sold to these businesses and built software for them for most of my working life, and I’ve seen every item on her list.
Her list wasn’t even complete. A customer with a small budget can still need a lot of support, a business can close for reasons no product can fix, and selling to thousands of individual owners costs money. Caring about these people doesn’t make any of that disappear. Small business has been a bad software market.
It was never a software market
It was never a software market
A one-person business has almost no software budget. It pays for labor instead, an accountant and maybe an assistant for a few hours a week. Whatever is left, the owner does herself in the evening. Alya pays her accountant about $235 a month. All her software together costs less than $50. The rest she pays for in evenings.
In the United States alone, about 30 million businesses have no employees. That’s close to four out of five of all businesses in the country. The number usually grows 2–3% a year. In the latest count it grew 5%. For thirty years the software industry looked at this market, saw the subscription line, which is small, and missed the labor line, meaning the money owners pay other people and the hours they put in themselves.
Software could only sell these people tools, and a tool leaves its owner with a job. She has to learn the interface, enter the information, decide what happens next, check the result and come back when something changes. Put five useful tools together and the owner gets a sixth job: keeping them consistent. In a larger company that work is shared, but in a one-person business it comes back to the same person.
Software sold to a solo business also costs the owner’s attention. Small checks are a fair price for something that saves 20 minutes and costs attention to run. Churn comes from the same place, because you abandon a tool the day you stop feeding it.
AI changes what can be sold. A system that knows the business, remembers what was agreed and finishes the task is a share of a job. And a share of a job is worth what the owner used to pay for it, in money or in evenings. That is a different number from the price of a subscription.
It’s a bet, and it has a ceiling. The average business in that count brings in roughly $60,000 a year, so nobody here will pay a salary for software. My bet starts with the owners who quote before they work, because a won proposal pays for the help that wrote it. Today Jinna is priced like software, by the month, and that is the right price for now: the work hasn’t yet earned a labor price. Willingness to pay more has to be demonstrated, one finished job at a time. But I’m confident of the frame. The investor was right that small business is a terrible software market. It was always a labor market.
Four departments missing
Every business, whatever it makes, runs the same loop: find the client, agree on the work, do it, bill it, get paid. The day I sold those 300 books, I ran all of it on foot. A large company hires a person for each step. A one-person business hires nobody and does all of them, after hours, and nobody is good at five jobs at once.
We call that a small business. It isn’t. It’s a full-sized company with four departments missing, staffed at night by one tired person.
Thirty million of these companies already exist. What’s new is that one person’s business can now be as complete as a staffed one. The agreement is kept, the invoice is right and the payment is chased, without the owner carrying every step in her head. One person still decides everything, and the rest is carried.
Companies grew, in large part, because functions needed heads. If functions stop needing heads, the natural size of a business becomes whatever its owner wants it to be. Some owners will want to grow. Many will want to stay one person, by choice, with help. Alya is happy behind the counter. She never wanted the admin. She wanted it gone.
The human was the integration layer
Why did single-purpose tools win for so long? Because context was free. The owner carried it. She remembered that this client agreed to $1,200 with work starting after the fifteenth, that he pays late, that the last invoice had a discount, and she kept all of it in her head, between the proposal app, the invoice app, the to-do list and email. Every specialist tool was built in isolation and assumed a human would glue it to the rest.
Consider a photographer who has agreed to shoot a wedding. There’s a date, a price, a deposit, a scope and a balance. The couple asks for an extra hour, and the photographer says yes. Now several things have to stay consistent. An AI can write a lovely reply and leave the old price in the proposal. It can issue the right invoice and then send a reminder after the money has arrived. Each output looks plausible while the job, as a whole, goes wrong.
An agent hasn’t taken work off your hands if you still have to keep it in your head.
The moment the owner stops carrying context, the old arrangement breaks. The context has to live somewhere, and wherever it lives becomes the real product. In Jinna it lives in a record of what was agreed. Once the photographer updates the proposal with the extra hour, the invoice for that wedding takes its amount from it, never from what the model happens to recall.
A proposal written by a system that remembers the last invoice beats one written by the best proposal tool on earth that doesn’t. That is my argument against five separate apps. There will still be excellent specialist tools, and an agent may use several. The boundary that matters is who carries the context between them, and putting five apps behind one login doesn’t move it. An all-in-one suite files the documents. The agreement keeps moving after they’re signed: an extra hour, a later date, a discount given on the phone. Someone has to hold the current version, and today that is usually the owner.
What it takes to carry a job
Cheap model calls can produce expensive supervision. If the owner has to check, repeat and repair, that time belongs in the cost of the work, and a system that needs constant watching hasn’t changed the economics at all. So everything above depends on reliability, and reliability turns out to be mostly about memory, permission and checking.
A business has facts, and it has promises. We agreed this price. We said Friday. This client has an exception. This invoice waits because the scope isn’t approved. Memory has to keep the difference between a passing suggestion and an agreement, between the usual policy and a specific exception, and it has to know which version is current. Otherwise remembering more just makes the system confidently wrong in more ways.
Then the memory has to reach the action. This one I learned the hard way. For a while Jinna could save a business’s rules, and the part that wrote documents never saw them. The memory lived in the chat, and a proposal could be drafted that ignored every rule the owner had just confirmed. Remembering is cheap. Remembering at the exact moment a document is written is the product.
It doesn’t take much. What Jinna puts in front of the model about a business is a card of roughly five hundred tokens, a few hundred words, a small fraction of what the model reads each time, and the work is in choosing them.
Permission has to be specific. Drafting a proposal, sending a routine reminder, changing a price and committing someone’s Saturday carry different consequences. Being right about the client’s name doesn’t earn the right to offer him a discount. We built Jinna never to save its own guess about a business without asking, and to show the owner everything it knows. Autonomy has to be earned one fact at a time, and the owner must always be able to see what happened and why, and change the instructions.
The system also has to check its own work before the owner sees it. Jinna re-reads the saved document before it quotes a figure from it. When a client’s name matches two invoices, it asks which one.
One more condition has nothing to do with reliability. The memory has to be portable. If the accumulated understanding of your own business is something you can’t inspect, correct or take with you, your independence has acquired a new dependency. In Jinna the owner can already see and correct all of it. Taking it with her is the part we still owe. What keeps a customer should be that the system knows the business, not that she can’t leave.
What we're building
Jinna is the brain and the hands of a one-person business. The brain is a memory of your business that you can see and correct: what you’ve agreed, what you charge, how you work, who owes you what. The hands are the parts of the loop it carries. Today that means proposals, estimates, invoices, the emails around them and the reminder when a payment is late. They are learned from the owner and from the business’s own website, written in the owner’s terms and sent when the owner says so. That isn’t Alya’s café yet. It’s the photographer, the architect, the consultant: anyone who quotes before they work.
That’s the starting point. More than 2,500 people have signed up, and nearly $5 million in proposals, estimates and invoices has run through Jinna. The version built around memory shipped this month. The end state is a system that runs the whole loop, from a client’s first contact to money in the bank, within whatever authority the owner chooses to hand over. I think that’s years away, not months.
If the whole loop works, Jinna is the back office for millions of companies in one country alone. The freedom to stay small belongs to their owners. The company that serves them won’t be small.
None of this is a free pass on the business model. We still have to reach customers at a sensible cost and earn weekly use. Taking on more of the work raises the price of getting it wrong. The questions I want answered by building and measuring are plain. What share of a job gets completed reliably within the owner’s instructions? How much attention do the exceptions consume? Do people come back in the ordinary week after the demo? The demo is the beginning. The ordinary week is the test.
Two things I don’t know. The first is how fast the models close the gap between “usually right” and “trusted with money”. My guess is sooner than the skeptics think and later than the demos suggest. The second is which parts of the second job stay human. My current answer is the promises. A system can draft the proposal. Only the owner can decide what she is willing to promise, and to whom.
Half a house, again
The computer I bought at 20 cost half a house, and it did almost nothing on its own. Everything it made, I made through it, after hours, the way owners of one-person businesses have worked ever since.
Today the raw capability costs about what one regular spends at Alya’s counter in a month. The system that turns it into a colleague, one that remembers, acts within its permissions and finishes, is what still has to be built. That’s the gap I’m working in.
People who start a business are betting on their freedom. For most of history the bet came with a second job attached, the one that ate the freedom it was supposed to buy. It won’t have to for much longer. People should be able to build something of their own and still have a life of their own.
I’m building for the day Alya locks up, goes home, and there’s nothing left to do.