I am in the car listing skills: marketing, programming, design, accounting, sales, market research. My point is that somebody who already knows how to do useful work may have more ways to offer it now. Add AI. Add a language that lets you work with customers elsewhere. Start looking beyond the market immediately around you.
That is the opportunity I mean. My claim that 2026 is the best year ever is enthusiasm, not a historical ranking I can prove. And “zero capital” is too loose: a service can need less upfront spending than a shop while still requiring equipment, connectivity, time and a way to reach customers.
Sell a result you can stand behind
A customer does not buy the fact that I used AI. They buy a working page, a clear design, useful research or some other agreed result. I still need enough knowledge to judge whether I delivered it.
Research gives a reason to investigate the tools without treating them as magic. Generative AI at Work reports a 15% average increase in issues resolved per hour among 5,172 customer-support agents, with effects that differed across workers.[1] That is evidence from a particular workplace, not a forecast of my business income.
A foreign language can help me understand the brief and explain the result. It does not remove the need to know the customer’s context. Automated translation is something to check, especially when a small wording error changes the offer.
Do the project maths before promising the margin
Here is a hypothetical €200 fixed-price project, not a quoted market rate or a result from the study. Suppose AI reduces production time but requires more checking:
| Work | Original workflow | Illustrative AI workflow |
|---|---|---|
| Production | 5 hours | 3 hours |
| Checking and revisions | 1 hour | 2 hours |
| Total | 6 hours | 5 hours |
| Gross revenue per hour | €33.33 | €40.00 |
The total time saving is 1 ÷ 6 = 16.7%. At the same fee, gross revenue per working hour rises by 20%, because €40 ÷ (€200 ÷ 6) − 1 = 0.20. These percentages differ because their denominators differ.
For 20 such paid projects, revenue is still 20 × €200 = €4,000 in either workflow. The assumed time falls from 120 to 100 hours. The 20 freed hours do not automatically create another customer. These figures exclude software, other expenses, taxes and unpaid sales work; they are not profit or take-home pay.
Make the first offer small enough to test
Choose one service you can already evaluate. Define the deliverable, what is included and how you will check it. Then try it with a real customer and record the actual time, including revisions. That result will tell you more than announcing that you run an AI business.
My piece about returning to Kosovo with a business plan asks the related question: what concrete work will support the idea?
The tool can help me move faster. I still have to build something somebody needs.
Sources and calculation notes
-
[1] Brynjolfsson, Li and Raymond, Generative AI at Work, arXiv version 2. Customer-support setting; the reported average is not a universal effect or a success probability.
-
Illustrative arithmetic: 5 + 1 = 6 hours; 3 + 2 = 5 hours; fee €200; 20 projects. Rates and percentages are rounded to two decimals and one decimal respectively. Reproducible inputs. No actual customer, price, cost quote or earnings outcome is represented.
-
Original video and thumbnail: Eduard Luta on TikTok. Published 2026-05-23. Research checked 27 September 2026.