OpenAI's Dot Personal Agent Generated 155 Activity Traces, 307,000 Tokens Unattended

An Agent That Works on Its Own

Security and governance vendor Geordie has been testing OpenAI's always-on personal agent dot since the company announced it in September. According to the firm, one dot agent produced 155 separate activity sequences and consumed more than 300,000 tokens in the course of a day, all without further human interaction.

Geordie chief AI officer Hanah-Marie Darley said dot "looked a little bit different" from other personal agents such as Meta's Muse and xAI's Grok Bot. The testing was deliberately limited: dot received only read-only permissions and was not connected to personal apps, email, or communications tools such as Slack. Even so, the agent quickly went into research mode, collecting as much information as possible and surveying the systems it could reach, then logging its findings in something akin to a daily journal that it updated throughout the day.

Two Agents, Constant Activity

According to Darley, dot always operates with at least two agents running. The primary agent interacts with the user, while a secondary agent continuously runs in the background collecting data. Dots and similar agents can also spin up short-lived subagents for specific tasks. In one trace Geordie analyzed, the agent spawned multiple subagents that checked all code repositories in a development system and continuously inventoried the data they found, none of it prompted by the user.

Of the 155 agent traces, at least 50 occurred within a two minute period, and the majority took place within a couple of hours. "The user would have no idea that there were sub-agents working on their behalf," Darley said. The 155 traces cost roughly 307,000 tokens in total, while individual requests ranged from a couple hundred to several thousand tokens depending on how much research dot performed. She called this "self-directed research," since the user never asked the agent to perform these tasks or to use a set number of tokens.

Token Spend and a Warning for Businesses

Token costs vary widely by location, subscription tier, discounted pricing, and computing costs, but Darley estimated that 200,000 tokens on OpenAI's Astra model cost between $5 and $10. Scaled to daily use, she said, that could mean spending what you pay for a latte every day in some cities, without any interaction from the user. She suspects many people are not prepared for the background spend of always-on agents, and noted that connecting token usage to actual work product is a subjective judgment.

Darley expects consumers to bring personal agents into work life, just as they did with chatbots and coding agents before them. Her advice to organizations is to try the technology themselves, since she believes agents can improve productivity, but to treat them with caution. "Don't give them any personal information that you wouldn't be comfortable sending in an email," she said, and for purchases, load a gift card rather than linking a real payment method, because people often do not think about that until something goes wrong.

Sources: Theregister