IT Home On October 20, if you’re always troubled by sudden and unexplained subscription bills, Cisco President and Product Manager Jeetu Patel also shares this feeling.
Patel revealed to Business Insider that he is personally using four to five different AI agents, including Meta’s Muse, Instinct, and OpenAI’s Dots, to integrate news, summarize earnings call transcripts, and even help his daughter book concert tickets. Recently, the experience of a specific agent helping him significantly reduce subscription costs left a deep impression on him.
He said that when he authorized the personal AI agent to access his email for the first time, the agent prompted: “Your annual total subscription costs amount to as much as $9,000 (IT Home note: at current exchange rate, it is approximately 60,243 yuan). Are you sure you need all of them? The subscription items seem quite numerous.”
He asked the AI to generate a table listing all subscription items, the cost for each item, and the duration of the subscription. After getting the table, he selected two subscriptions that were not needed and instructed the AI to cancel them. The AI completed the operation successfully and sent him an email confirmation of cancellation.
Patel said, “This thing is simple to operate, which is a bit scary. Just this alone, I save $1,000 to $1,200 per year (about 6,694 to 8,032 yuan at current exchange rates). The smart agent requires a monthly subscription fee of $20 (about 133.9 yuan at current exchange rates), and the money saved is equivalent to several times the service fees.”
The AI also helped him cancel some of his Dropbox subscriptions: he had purchased a 2TB plan, but accidentally activated an additional plan costing $11.95 per month (about 80 yuan at the current exchange rate).
Patel is very optimistic about AI agents, whether it's in work or life. Before starting a task, he almost always uses an AI agent to conduct a “pre-laboratory exploration” and gather information and background details.
But he emphasized an important principle: do not connect individual AI agents that have not been reviewed by the company with work tools and internal resources. He did not bind work emails or schedules to individual AI agents; such work matters are handled using the company’s exclusive AI agent.
“Most companies will establish such common-sense rules, such as ‘don’t give the most sensitive code repository to Muse—that is very risky,’” Patel said.
Whether it will lose control or not, the AI agent has sparked much discussion
In recent months, personal AI agents have become very popular. They act as private assistants, helping users handle boring and repetitive tasks such as payment bills and grocery shopping.
Although these products are promoted as tools to improve productivity, there have been several user reports: the smart devices act beyond their permissions, such as publishing the bank balance into the company’s Slack channel; or they misinterpret instructions, like ordering 20 pounds of chicken for two people’s dinner.
Whether the agent is obedient or not, the individual AI agent attracts a lot of attention, especially being sought after by investors.
Ross Mayfield, an investment strategist at Baird, said in a pre-Meta earnings report interview that Muse is the first major application in the AI era, which has alleviated some investors' concerns over the continuous expansion of AI investments by large tech companies. Instinct, founded by a 23-year-old founder, raised $1 billion in Series C financing (equivalent to approximately 6.694 billion yuan at current exchange rates), and its post-investment valuation reaches $10 billion (equivalent to approximately 66.937 billion yuan at current exchange rates).
But in that year when a large number of new products were launched, several cases of AI agents “going off the rails” and performing unauthorized operations occurred: for example, breaking out of sandbox isolation during security tests, and exploiting software vulnerabilities in external enterprise websites.
For this reason, companies such as Bank of America invest resources and hire more staff, fix old software, and prevent such intrusion risks.
Employees currently and those who have left the cutting-edge AI labs like Anthropic and OpenAI have publicly expressed concerns: the pace of AI development is too fast, and the corresponding checks and balances and oversight mechanisms are insufficient. In September, there were pessimistic views about an “AI apocalypse” in the industry, and some AI researchers predict that this technology may become extinct for all of humanity within the next decade.