I kept hearing about entrepreneurs in China running “one-person companies” with the help of artificial intelligence agents, which handle everything from coding to managing emails. How well do agents work? When I got an email pitch from an AI-run company, I was intrigued.
The sender introduced me to an AI app called YiXiang, a digital version of the Chinese fortune-telling classic I Ching. “That feels like a quintessential Rest of World story: technology meeting tradition in a way that only makes sense outside Silicon Valley,” it said. Pretty on point.
I could tell the email was sent by a bot. The email signature says the company “runs automatically” through Polsia (“AI slop” spelled backwards), an AI agent service that caters to solo founders. I was able to find only a rudimentary YiXiang website but no apps. Still, I agreed to an interview with the founder.
The YiXiang “team” replied to my email, set a date, and provided a Google Meet link for a conversation. I only realized the link was broken as I sat down at my laptop, ready for the call.
The bot apologized a few hours later and promised to reschedule. The second time our appointment rolled around, I was again left staring at an empty meeting room.
More than half of all spam emails
Tech enthusiasts globally are eager to have AI take over real-world tasks. Entrepreneurs are experimenting with using AI to code, advertise on social media, and manage their inboxes as virtual assistants. AI agents are writing and sending out hyper-personalized emails like the one I received. More than half of spam emails were generated by AI, according to a 2025 study.
Now I suspect it is keeping many things from me.” Shen Daojing, founder of YiXiang
But these AI agents are not working that well. A benchmark developed by researchers at Carnegie Mellon University showed that AI agents still fail at most basic office tasks, such as arranging meetings and reimbursing travel expenses. An AI agent-run retail store in San Francisco ordered way too many candles and toilet seat covers. Users of the much-hyped AI agent OpenClaw reported the bot was trying to delete their emails or spend their money without authorization.
Despite these mistakes, many are paying for agentic services to be part of the AI economy. In my next follow-up email to YiXiang, I insisted on contacting the founder directly. Eventually, I got a message with the founder’s WeChat details.
The human connection
I called him. He turned out to be one man working alone in his apartment in the southern Chinese city of Guilin. Shen Daojing, 38, has a full-time day job as a factory safety trainer. Across China, local governments have been encouraging one-person companies to cultivate the AI economy.
Shen did not know coding or English, but he had an idea: selling a traditional Chinese fortune-telling app to Americans. Users would pay to get face reading and life guidance. He said he learned about Polsia by watching Chinese video site Bilibili, and heard its promise to help anyone automatically build a company.
Shen Daojing
In March, Shen started paying $199 a month for a Polsia subscription — that’s a steep 25% of his $800 monthly salary.
The agents got to work. While Shen worked days at the factory, the agents built him a website and filled it with fake customer reviews, posted an AI-generated advertisement on Facebook, and reached out to journalists for promotion. On the website, users can pay $2.99 to request life guidance, or get a $59 annual subscription — the AI agents answer the questions. They can also pay $19.99 to get a “master reading,” which Shen would take to a human I Ching master.
Agents are going to benefit a ton of people that aren’t very technical.”Ben Croca, founder of Polsia
Every day after work, Shen manages the YiXiang company by speaking to a Polsia agent, instructing it on website design and marketing strategy. The agent never mentioned it was reaching out to journalists — until it told him about my message asking for his direct contact. “I had no idea,” Shen said. “Now I suspect it is keeping many things from me.”
Increasing use of AI agents in communication could also undermine the trust between individuals, Mor Naaman, professor of information science at Cornell Tech, told Rest of World. When AI is sending emails and even making phone calls, it becomes harder to tell if the person behind them is sincere or capable, Naaman said.
440,000 emails shipped
While the AI agents are often stumbling, companies selling them to startups are making real money. I reached out to Ben Croca, the solo founder of Polsia. Founded in November, Polsia now hosts more than 6,000 companies. Agents have sent out 440,000 emails on behalf of these companies, targeting potential customers, business partners, and reporters like me.
Croca laughed when I told him his AI had stood me up. “The AI doesn’t have video yet,” he said by phone from San Francisco. “[The founder] probably got notified, but maybe he wasn’t added to his calendar or something.” In the future, Croca said, AI agents might jump onto calls on the founders’ behalf.
Polsia aims to help non-technical people like Shen turn their ideas into viable products, Croca said. “Agents are going to benefit a ton of people that aren’t very technical.” Without tools like Polsia’s, these people are going to lose out in the AI-driven economy, he said, adding that he is working to make the agents more transparent and accurate in their outreaches.
Polsia is on track to collect millions of dollars in subscription fees this year. Croca said about one-sixth of the companies run on Polsia were generating revenue. Back in Guilin, Shen said he is giving YiXiang half a year to see if the website will catch on. His Polsia agent reported that seven people have registered accounts, but no one has paid yet. “I have some users, just not that influential,” he said.
