At a recent corporate event, OpenAI's CEO Sam Altman made his first systematic proposal of the three-phase theory of AI product development and clearly stated that the next breakthrough in the industry will be "Proactive AI" - an AI system that runs continuously in the background and actively assists users.

Altman divided the evolution of current AI products into three clear stages. The first stage is chatbots represented by ChatGPT, where users need to initiate a conversation to get a response; the second stage is AI agent systems with task execution capabilities, such as OpenAI's Codex, which can independently complete specific tasks like programming; and the third stage, which Altman believes is the most promising direction, is more automated "Proactive AI." He said directly, "I bet the next thing that comes out will be this continuously running Proactive AI. If there's anything worth preparing for in the next year, it's this direction I'm most optimistic about."

image.png

Although Altman referred to the current agent stage as the "most popular AI product category so far," and its development has largely been driven by the actual needs of enterprise customers, as the product lineup continues to expand, users have encountered new confusion. Many enterprise users are unsure when to use chatbots, when to call upon Codex or APIs, and they also struggle to integrate the required contextual information and various plug-in tools. This product fragmentation issue has prompted OpenAI to begin thinking about deeper integration solutions.

To address this, Altman revealed that OpenAI is developing an integrated product, planning to deeply merge Codex's agent capabilities with ChatGPT and other tools, forming a unified platform similar to a "super app." This platform is seen as the infrastructure for "doing all the other things we want to do," aiming to solve the current problems of scattered products and high user learning costs.

At the same time, AI cost issues are rising from peripheral topics to core considerations in business decisions. Altman pointed out that at the beginning of the year, companies rarely proactively discussed costs, but now it has become a "huge issue" and may be "the second most important topic." He cited the example of Uber, which exhausted its entire AI budget in just the first quarter. In response, OpenAI's strategy is to improve model efficiency to reduce costs while creating more value for customers. "I think we will have many ways to help our customers get more value with less investment."

A deeper challenge lies in user awareness. Altman admitted, "Most people" don't know how to use AI efficiently. Many have realized they haven't utilized AI sufficiently and haven't fully leveraged its value, but learning new working methods itself has a barrier. "It's not easy to learn new working methods, and the start-up cost is a bit high." Even if OpenAI can demonstrate many complex and powerful application cases, most clients won't use AI in that way in their actual business.

"Proactive AI" is OpenAI's response to this practical dilemma. If users are unwilling to actively learn how to use AI, future AI systems should be integrated into work scenarios, run in the background, and automatically complete tasks around users. Altman's vision is that users may not even need to understand what AI can do. "Can OpenAI function as an agent running in the background, connecting all the context of my company? Don't make me try to understand what it can do, just make it useful for me."

However, this vision also poses new challenges for existing IT architectures. Unlike traditional chat windows that respond on demand, an AI system that runs continuously and accesses a large amount of enterprise contextual data raises higher requirements for data security, permission control, and computing resource allocation mechanisms. Enterprises need to redesign AI deployment methods, security strategies, and infrastructure to adapt to this "always-on" intelligent form.