Young Businesses and AI
Something changed in the last eighteen months in the way a young business can be built, and it has quietly returned the advantage to the way your parents worked.
Before we start, let me say that intellectual capital is now commoditised with AI, and so this piece is for those of you who are building an actual product or service. An actual product is something that can be seen with eyes and used with hands, and a service is an experience that is felt either within an environment that you create, in a physical space or virtually.
You already know your numbers.
The reel went up and by the end of the second day it had a few hundred thousand views. Around thirty people commented. Someone with 40k followers shared it, which was the plan. Then you opened the dashboard later in the week and there were four signups, and one of them was your cousin.
If you are the mother or father of that founder, you have watched this from the other side of the room. Your child looks at a phone for eleven hours and calls it work. Some days there is great excitement over a number that does not appear to be money. You have not asked too many questions, partly out of concern and partly because you were not sure you would understand the answer.
The answer is simpler than it looks, and the news is that it is better for your child than the last two years have felt.
The method that worked, and stopped
For about ten to twelve years, a certain method genuinely worked.
You made something. You made videos about the thing, you talked about it and you paid Instagram or Google a small amount to show those videos to strangers. Some of those strangers became customers, and if each customer was worth more than what you paid to reach them, you had a business. It was not easy, and it was not guaranteed, but it worked often enough that an entire generation of young people learned to build companies this way. Your child did not invent this approach. It was taught, loudly, by very successful people, and it was correct at the time.
That method has stopped paying. The people still teaching it formed their opinions in the years when it worked, which is why they still sound so certain.
Start with the price of a customer.
Two years ago an Indian brand selling online could reach one new customer for around ₹300. Today the same customer costs ₹500 or more, and in categories like beauty the figure sits between ₹800 and ₹1,200. On Instagram and Facebook the cost went from roughly ₹380 last year to ₹502 this year, a rise of thirty-two per cent in a single year. Across categories it is up thirty-five per cent.
The reason is an auction. Nobody at Instagram decides the price. Whenever there is space to show an advertisement, thousands of businesses bid for it in the time it takes to blink, and the highest bid wins. What has changed is that most of those bids are now placed by software that never sleeps, never gets tired and never decides it has spent enough this month. Every business is bidding better than it used to. So the price goes up for all of them.
There is a second number that decides whether a business can exist at all. To survive now, a brand has to earn back close to four times what it spent to get a customer, which it can only manage if the same person comes back and buys two or three more times. One sale is no longer enough to cover the cost of finding the person who made it.
The free route has closed too, and this one matters more.
For years, if you could not afford advertisements, you could write. You wrote something useful, people searched for that thing on Google, they found you, and some of them stayed. It was slow and it was free and it built many good businesses.
Now, when you ask Google a question, Google often answers it directly at the top of the page and you never click anything. Somewhere between sixty and sixty-nine per cent of searches end that way. In Google’s newer AI mode it is ninety-three per cent. Across 2025 and in the first six months of 2026, the traffic Google sent to websites fell by about a third. One British newspaper measured what happens when Google’s own answer appears above its link: the share of people who clicked through went from twenty-five in a hundred to under three in a hundred.
The road is still there. Almost nobody walks down it any more.
The edge that evaporated
Making things that persuade people used to be expensive.
A page that convinced a stranger to buy needed somebody who could write. A launch video took a week and a person with taste. If you had that skill, the shop next door did not, and that difference was worth real money. Every book about growing a business in the last decade is, underneath, a book about how to use that difference.
Between the end of 2022 and late 2024 the cost of using these AI systems fell more than two hundred and eighty times over. Between late 2024 and now, it fell by more than this number again.
Everyone now has the same machine. Your competitor has it. A sixteen-year-old in Poland has it. The well-funded company with forty employees has it as well, plus money. Whatever edge existed in making convincing things for a screen was an edge in a world where convincing things were rare, and they are not rare any more.
The results are visible. Of the young companies built on top of these AI tools, somewhere between eighty and ninety-five out of every hundred fail. Sixty to seventy never earn a single rupee. Three to five in a hundred reach ten thousand dollars a month. Most of those founders worked hard. Most of them posted every day. They did exactly what they were told to do.
Uber, told properly
Uber is worth looking at, because its story is usually told wrongly.
Uber is remembered for speed and aggression and money. What its early years actually consisted of was one person arriving in a new city with a laptop and a local phone number, walking to the parking lots where taxi drivers waited between shifts, and signing them up one by one. Standing outside airports. Paying drivers to keep circling empty streets so that the map would not look dead when the first customer opened the app. City after city, in the heat, arguing with unions and transport departments and sometimes the police.
There was no video in any of it. The people went first and the technology came afterwards, and the technology was good because it was built by people who had stood on a road at two in the morning in a city where nobody could get home.
That kind of hard work has not stopped paying. It is the only kind that still does.
Two roads, and what the money is actually buying
Two roads are usually put in front of a young founder.
The first is to take one small piece of a big mess, build a neat tool for that piece, and rent it out every month. Most of the software companies you have heard of were built this way, and it was an excellent idea for twenty years.
The second is what the large investors are doing right now, and it is close to the opposite. Instead of selling software to a company, they buy the company. One American investment firm has put a billion and a half dollars into buying accounting practices, call centres, property managers and computer service firms, and then rebuilding how those businesses work from the inside. The arithmetic is plain. In a service business, more than half of every rupee goes on salaries. If a machine can take over a good part of that work, each person handles two or three times as much, and an industry that has looked the same for a hundred years changes inside eighteen months.
What that billion and a half is really buying is rooms full of people doing work, so that somebody can stand in the room and watch the work being done.
Which brings us to the strangest job in the world at the moment.
In 2005 an American company called Palantir invented a role for itself. Rather than sell software to a customer and leave, it sent an engineer to go and sit inside the customer’s office, at the next desk, for months, building the thing in front of them while they worked. For twenty years the rest of the industry thought this was a wasteful way to run a company.
This year everyone wants it. Anthropic has put one and a half billion dollars behind the idea, OpenAI something close to ten billion, and several of the large consulting firms have set up formal programmes. A survey of fifteen hundred of these engineers found they spend forty-seven per cent of the working week sitting with customers and thirty-one per cent actually writing software. They are paid between two hundred thousand and seven hundred and eighty-five thousand dollars a year.
And a firm that searches for such people estimates there are about two thousand in America who can do it. Not two thousand looking for work. Two thousand who exist.
The richest industry in history, holding machines that can write most of the software the world needs, has discovered that the thing it cannot get enough of is human beings willing to sit beside another human being and watch them work.
Stand somewhere specific for a long time
So the advice for a twenty-six-year-old with an idea, eleven thousand rupees a month for expenses and nobody’s attention is not the advice of five years ago.
Working harder at the computer will not help, because the computer is now equally good for everybody. Posting more will not help, because the feed has no shortage of things to show. What is still difficult, and still rare, and still entirely available to somebody with no money, is standing somewhere specific for a long time.
If you are building something for chemists, go and stand behind the counter of a chemist’s shop in Nagpur for three weeks. Not a phone call. Not a form. From the time the shutter goes up to the time it comes down, watching everything that happens between a customer walking in and walking out, and noticing the one moment that makes the chemist’s shoulders tighten. If you are building for truck owners, ride in the truck. If you are building for doctors in a small town, sit in the waiting room for four days and count the people who leave without being seen.
You will come back knowing things that are written down nowhere. The shortcut everybody uses and nobody admits to. The form that is always filled in wrong. The software they already paid for, sitting open in another window, untouched since March. The exact moment on a Tuesday afternoon when a sensible adult decides the whole thing is not worth the trouble.
None of that can be produced by a machine, because none of it has ever been written down.
The part I would like the parents to read twice
What I have just described is how you worked.
If you ran a shop, you knew which customer’s daughter was getting married and what her mother would come looking for in October. If you taught, you knew which child could not see the blackboard before that child knew it. If you sold anything, to anybody, you did it by being physically present in front of them for years, and you learned their lives, and the business came out of that knowledge.
For about twenty years your children were told that this was slow and old-fashioned, that the future belonged to whoever could reach ten lakh strangers from a bedroom. For a while that was even true.
It has stopped being true, and the most advanced companies on earth are now paying enormous sums to relearn what you did without a name for it.
The pressure on your child while all this settles is real, and it deserves a number too.
Entry-level jobs in Indian technology companies fell forty-four per cent in a single year. Roles for freshers have gone from about six lakh at their peak in 2022 to the lowest on record. Fifty-five per cent of Indian IT firms cut their entry-level hiring, against fourteen per cent who cut senior roles, which tells you exactly which rung of the ladder was taken away.
They stopped hiring young people because the work a company used to hand a beginner is work a machine now does. That same change has raised the value of anything your child knows that a machine does not. A machine has read everything ever written about chemist shops and has never once stood in one at seven in the evening on the day before Diwali.
Karuna
All of it ends in the same place. You sit with a person long enough to feel what their day costs them, and then you build something for it.
There is a word for that movement and it is karuna. It is usually translated as compassion, which in English has picked up a flavour of charity, of the comfortable person feeling something on behalf of the uncomfortable one. The Sanskrit means something firmer. Karuna begins with recognition, with the distance collapsing between your situation and theirs, and it rests on understanding that the trouble in front of you is the same kind of trouble as your own, happening to the same kind of creature. It is Vasudhaiva Kutumbakam applied to a customer.
When a company pays seven hundred and eighty-five thousand dollars for somebody to go and sit with a customer, that is what it is buying. Not intelligence. Not output. The ability of one person to be affected by another person’s difficulty, accurately, from close by, for long enough to do something about it.
That ability was always there. For about a decade it was also optional, because an Instagram advertisement could stand in for it and carry you a surprisingly long way. That is the honest reason so many of us learned to run businesses without ever standing in the room.
It has stopped standing in. Everything that came after it is now free, and therefore worth nothing.
The only miracle in the world is you standing on your own two feet.
Hari Om Tat Sat.
Arjun is a 30-year practitioner in the Bihar School of Yoga tradition. He founded OMJOOMSUH in 2022.
Related
Karuna · Vasudhaiva Kutumbakam · Yogic UX · Four Dharmas · Sanatan Dharma · The Three 3.0s · No Bank Would Lend His Father Two Lakh Rupees · The Four Rooms Of Talent · Of All The Things, He Chose That One
Sources
- Indian cost of acquiring a customer by category, the move from ₹380 to ₹502 on Meta, the thirty-five per cent rise and the four-times earn-back rule: Reduce CAC for Ecommerce India 2026; Why Your D2C Brand’s CAC Keeps Rising, Adtitude Media.
- Searches that end without a click, Google’s AI mode, and the newspaper click-through collapse: SparkToro, 2026; Search Engine Journal on AI Overviews and publishers.
- The fall in the cost of using AI systems, and the survival rates of companies built on them: Proprietary Data Moats and AI Startup Defensibility in 2026; Are AI Wrapper Startups Worth Building in 2026.
- The engineers sent to sit with customers, the survey of 1,500, their pay, and the estimate of 2,000: TechCrunch, July 2026; State of Forward Deployed Engineering 2026.
- Investors buying service businesses outright, and the salary arithmetic: Inside the VC Roll-up Craze, Newcomer; AI Rollups Could Eat the World.
- Indian entry-level hiring: The Death of Mass Hiring, OwnYourCareer; India’s Tech Hiring Slowdown 2026.
- Andrew Chen, The Cold Start Problem, 2021, for how Uber opened its early cities.