OpenAI is preparing to unveil an AI agent, anticipated to launch in January, as per a report from *Bloomberg*. Dubbed “Operator,” this agent will possess the capability to take control of a user’s computer, executing tasks like booking flights or writing code.

During a Reddit AMA (Ask Me Anything) session alongside fellow executives, OpenAI CEO Sam Altman hinted at this advancement. Altman remarked, “the next major breakthrough will be agents.” Chief Product Officer Kevin Weil also highlighted that a major feature for 2025 will be ChatGPT’s capacity to start conversations with users. This comes on the heels of reports from September, when some users observed ChatGPT initiating messages, which OpenAI had clarified was unintentional at the time. Nonetheless, it may indicate prospective functions.

AI agents are increasingly becoming a prominent focus within the technology sector. For example, Microsoft has incorporated AI agents within its Copilot model, enabling organizations to tailor them for automating tasks. Similarly, Anthropic has rolled out a capability for its Claude model that can manipulate a user’s cursor and write code. Additionally, there are speculations that Google is working on a similar tool, codenamed “Jarvis,” designed to assist users with web browsing, shopping, and flight bookings. A leaked document indicated that this tool could be launched as a Chrome extension.

Nonetheless, recent insights from *Bloomberg* and *The Information* indicate that large language models (LLMs) are encountering developmental hurdles. In spite of enhanced computing capabilities, enhancements in these models are becoming less significant, leading to diminishing returns due to fundamental limitations in generative AI architecture. This scaling is also associated with substantial costs.

AI specialist Gary Marcus, who forecasted these limitations back in 2022, remarked that the high valuations of firms like OpenAI and Microsoft are predicated on the assumption that LLMs will eventually transition into artificial general intelligence (AGI). However, Marcus contends that there is no definitive resolution to problems such as AI “hallucinations”—where models produce inaccurate or nonsensical outputs—without systems that can explicitly represent and reason about facts.

In spite of these issues, Altman remains hopeful, asserting in the AMA that AGI is “attainable with current hardware.” Meanwhile, OpenAI continues to refine its existing models and introduce new features, primarily building upon variations of its current LLM technology.