Search for how to make money with artificial intelligence and every result tells you the same thing in a different order: the margin is enormous. AI product photography costs a fraction of a studio. AI video costs a fraction of a creator. Automation costs a fraction of a hire.
All of that is true. None of it is the reason most people never get there.
A 2026 survey of more than a thousand working adults put the realistic income from AI-based side work in the first six months at 500 to 1,000 dollars a month. Not per day, which is what the screenshots claim. Per month, and that is the median outcome for people who actually started.
So the interesting question is not how large the margin is. It is what ten thousand a month actually requires, in units you can count. This is that arithmetic for five businesses, with the division done.
1. Product photography
**The economics.** AI-generated product imagery runs about 0.35 to 2.50 dollars per image, and closer to 0.50 to 1.50 on a subscription once batch processing is included. Studio photography runs 25 to 200 dollars per image, and the quoted rate understates it: once retouching, studio rental, shipping and coordination are added, a 40-dollar quote lands nearer 84. For a 500-SKU catalogue the difference is roughly 12,500 dollars a year against under 1,000.
**What ten thousand a month requires.** Charge 12 dollars an image, which is a quarter of the cheap end of studio work and still a comfortable price for the buyer. Ten thousand divided by twelve is 833 images every month. At around a hundred images per catalogue that is eight clients, delivering continuously, not once.
**What is actually scarce.** Not the ability to generate an image. Access to eight e-commerce brands who will pay a monthly invoice. If you cannot list eight such businesses you could email tomorrow, the margin does nothing for you.
2. UGC and avatar video ads
**The economics.** AI-generated ad creative costs roughly 2 to 20 dollars per video against 50 to 500-plus for a human creator. Tool costs are trivial next to that: HeyGen from 29 dollars a month, Creatify and MakeUGC around 39, Arcads at 110 for ten videos, which works out near 11 dollars each.
**What ten thousand a month requires.** Direct-to-consumer brands burn through creative because performance decays with repetition, so this sells as a monthly quota rather than per asset. At 500 dollars a month per brand you need twenty brands. At 2,000 you need five. The five-brand version is a far better business and a far harder sale.
**What is actually scarce.** Knowing which twenty brands are currently spending on paid social, and getting to the person who owns that budget. The generation is the easy part and everyone has it.
3. Automation agency
**The economics.** Agencies building workflows on tools like n8n and Make charge monthly retainers of 1,000 to 5,000 dollars, with n8n-specific work commonly landing between 500 and 3,000 per client. Published playbooks describe five clients at 5,000 producing 40,000 a month, and a twelve-month path to 50,000 in monthly recurring revenue with fifteen clients.
**What ten thousand a month requires.** Four clients at 2,500, or ten at 1,000. This is the highest revenue per client of the five and the smallest number of relationships to manage.
**What is actually scarce.** The ability to sell a 2,500-dollar monthly retainer to a business that has never bought automation before. That is a sales skill, it takes months to develop, and no amount of technical competence substitutes for it. The compensation is that retention here is the best on this list: once a company's operations run through your workflows, leaving is a project rather than a decision.
4. Faceless content channels
**The economics.** Revenue is views multiplied by RPM, and RPM is decided almost entirely by niche. Finance and investing pays roughly 12 to 25 dollars per thousand views. Technology sits at 8 to 15. Horror and true crime at 4 to 8. Motivation at 3 to 6.
That spread is the whole business. A hundred thousand finance views can out-earn two million gaming views, and the choice is made before the first video.
**What ten thousand a month requires.** At an 18-dollar finance RPM, ten thousand dollars is about 555,000 views every month, sustained. At a motivation RPM of 5, the same income needs two million.
**What is actually scarce.** Half a million monthly views. Nothing else on this list has a requirement that brutal, and nothing else takes as long: published timelines show months one to three producing essentially nothing, months three to six reaching 100 to 500 a month, and months six to twelve arriving at 500 to 3,000. The genuinely successful version reaches 3,000 to 10,000 within a year.
The failure case is worth more than the success case here. One documented channel with 2,162 subscribers and twelve videos over roughly two years earned 895 dollars in total. That is not a warning about doing it badly. That is what doing it occasionally looks like.
5. A product with AI inside it
**The economics.** Ten thousand a month is 120,000 a year in recurring revenue. That is roughly 345 customers at 29 dollars a month, 35 at 290, or 4 at 2,500.
Those are not three versions of one business. The first is a marketing company that needs reach and survives casual churn. The third is a sales company where four relationships fit in one person's head. Most founders end up in the first shape by accident, because consumer pricing is the pricing they have seen, and then discover they have built something requiring a marketing budget they never planned.
**What is actually scarce.** A reason the customer cannot leave. If the product is a capability plus an interface, the platform underneath ships it as a feature and the business ends in a release note.
The pattern across all five
Every one of these has a cost advantage between 80 and 99 per cent, and the median entrant still makes 500 to 1,000 dollars a month after six months of trying.
The two facts are not in tension. They are the same fact. Cheap production stopped being scarce at the exact moment it became available to everyone, and scarcity moved somewhere else:
In photography it moved to access to eight e-commerce brands. In video, to knowing who is currently spending on paid social. In automation, to closing a 2,500-dollar retainer. In content, to half a million monthly views. In product, to owning something the platform will not rebuild.
Not one of those is produced by a tool. That is the entire gap between the margin everyone quotes and the income almost nobody reaches.
How people actually fail at this
The documented failure pattern is specific enough to be useful. Someone picks three of the five businesses above, builds half of each, finds no client within thirty days, and concludes that AI side businesses do not work.
The conclusion is wrong and the behaviour is understandable. All five look equally viable from outside, because from outside they are all "use AI to do a thing cheaply". The differences only appear when you do the division, and the division is what nobody publishes.
One business, one requirement, counted. Eight clients, or twenty brands, or four retainers, or 555,000 views. Then ask whether you can plausibly get that in a year, and if the honest answer is no, pick a different one before spending the year rather than after.
What we do, and why the same lesson applies
We build AI trading agents and the terminals they run in, which is a sixth version of the same story. The models are available to everyone, the capability is not scarce, and the thing that is scarce is a record nobody can edit after the fact.
So every forecast we publish is serialised, hashed and timestamped into a Bitcoin block before publication, then scored in public with the misses on the same page as the hits. When our stated 50% intervals turned out to be containing 84% of outcomes, that is where we found it, and we published the diagnosis before we had the fix.
The general point holds outside forecasting. When a capability becomes universally available, the advantage moves to whatever the capability cannot supply, and the businesses worth starting are the ones where you already have some of that other thing.
Educational content - not financial advice.