Artificial Intelligence

One Person, 94K in Monthly Revenue: The Real Cost of Running an AI Business Solo

By 老金带你玩AI · Oct 31, 2026

Read original on juejin.cn ↗ Google Translate ↗ Alt translation

As AI makes "one-person companies" viable, the bottleneck shifts from capability to delivery discipline. This account spells out the mundane economics—commissions, cash flow, rework hours, scope creep—that determine whether a solo AI business actually survives, not just whether it generates a big revenue screenshot.

Summary

A solo operator in China, formerly a game developer for 15 years, documents his October 2026 revenue of 93,876.60 yuan from three businesses: paid courses, sponsored content, and custom software delivery. He's clear that this is gross revenue, not net profit, and the core of the post is about what that number doesn't show.

The article walks through the operational reality of each business line. Courses demand real teaching (guiding students until they can apply the skill on new materials), not just dumping resources. Brand placements require matching a defined outcome (exposure, clicks, registrations, or purchases) to the content, not just delivering an article. Software work requires acceptance by actual users, not just a demo. All three put the solo operator into sales, project management, support, finance, and QA simultaneously.

The post also covers the economics of solo delivery: channel commissions that can erase margins, the difference between high-revenue and high-value orders (a 20k order that eats 80 hours vs. a 6k order that takes 10), cash flow timing mismatches, and the true cost of AI tools when rework is figured in. A recurring theme is that scale requires systematizing repeatable steps, not just working more hours.

Takeaways

— Revenue of 93,876.60 yuan in October 2026 came from three business lines: courses, brand-sponsored content, and custom software delivery. It is gross revenue, not net profit.

— For courses, the standard is that students can apply the skill to a new, similar material. Guided practice, checklists, and independent exercises beat just handing over materials.

— For brand placements, define the target outcome (exposure, reads, clicks, registrations, actual usage, or sales) before starting, and split posts, revision rounds, publish times, and material usage.

— For software delivery, acceptance must be done by the actual users with real or authorized data, following their workflow. A smooth demo isn't acceptance.

— A solo operator still carries eight roles: direction/product, content/acquisition, sales/business, project coordination, production/quality, after-sales, finance/admin, and review/collaboration.

— Channel commission math example: a 1,000-yuan product with 700-yuan commission and 250-yuan delivery cost leaves 50 yuan before fixed expenses and taxes. With 800-yuan commission, each order loses money.

— A 20,000-yuan order with 4,000 direct cost and 80 hours yields 200 yuan/hour. A 6,000-yuan order with 1,000 direct cost and 10 hours yields 500 yuan/hour. Higher revenue doesn't mean better per-hour value.

— Cash flow trap: a 2,000-yuan fixed expense due in week two can come before a 16,000-yuan final payment in week three, even when the project has margin. Schedule all cash inflows and outflows.

— AI tool cost example: a 100-yuan call fee plus 10 hours of rework (at 100 yuan/hour) totals 1,100 yuan for 100 qualified results; a 300-yuan fee plus 2 hours of rework totals 500 yuan. cheaper call fees can cost much more in time.

— To productize, fix material format, delivery result, and acceptance criteria, then offer it as a standard service. You don't need subscription software.

— After each order, leave reusable assets: checklists, modules, test samples, and common error logs. Reusable is the method, not the previous answer.

Conclusions

The biggest surprise isn't the revenue but the insistence on teaching as a service. Simply selling a course without ensuring students can transfer the skill creates refund requests and reputation damage, which is more expensive than the time spent grading.

The commission example shows why many AI entrepreneurs end up losing money on volume: they focus on top-line revenue while ignoring refund rates, delivery cost per order, and the fact that channel partners may not understand the service burden behind the product.

The recurring theme of "rework time" across AI tools and outsourcing suggests that measuring real end-to-end cost (including your own hours) is essential. Many solo operators underprice because they forget their own labor has an opportunity cost.

The eight-role breakdown makes clear that a one-person business isn't a smaller company; it's a full company where every function is time-sliced. This explains why solo operators burn out when revenue grows: they add volume without reducing per-task overhead.

The advice to test AI tools on real, messy tasks (not clean samples) is a practical antidote to vendor demos, and applies to any developer evaluating AI-assisted workflows.

Source: juejin.cn ↗ Google Translate ↗ Backup ↗