We stand at a critical juncture of technological transformation, with Agents (intelligent agents) at the heart of the current AI frenzy, burdened with immense expectations. Yet, what we currently understand and painstakingly build as Agents may merely be a stepping stone on the path to the future.
Agents are the future, but the Agents of the future are not the Agents of today. To put it more bluntly, today’s Agents—the code we meticulously develop—are often just free training data for the next generation of models. One might even say this is a rapidly narrowing detour. As large models scale exponentially, the processes themselves become the data that models learn from and output. Once automation is internalized within the models, the very notion of Agents scheduling or being scheduled will become superfluous.
Humanity often grapples with new paradigms using outdated cognitive frameworks. The current fervor around Agents might simply reflect this cognitive lag. We persist in attempting to tackle a complex, dynamic real world with the intricate processes of traditional software engineering. This fundamentally contradicts the true essence of AI. The genuine power of artificial intelligence lies in solving systemic problems with simplicity, not complexity.
In the AI era, an Agent isn’t an App. Generative AI, in a sense, functions more like an intelligent vector database. From this perspective, the quality and value of data, along with how it’s leveraged to drive emergent intelligence, are what will truly define the future of the AI world. The key to this new world isn’t processes, nor even algorithms, but data. Data isn’t just valuable; it’s a high-value asset, yet it’s often being acquired for free. This is precisely the bottleneck impeding data flow, and it’s the critical area where we must delve deeper and achieve breakthroughs. Indeed, data value is the true key to unlocking the future of AI.
Furthermore, the core definition of a true Agent lies in its ability to alter its environment. This isn’t confined to digital space; it must exert tangible influence on the physical world. Consequently, a vast number of future Agents will deeply integrate with hardware. Without the capacity to interact with the physical world, an Agent’s value will significantly diminish, rendering it an abstract concept. Those Agents genuinely capable of reshaping the business landscape will be the product of an organic fusion of intelligent algorithms, massive datasets, and real-world execution capabilities. Clearly, this is far more challenging than mere code assembly.
But what appears to be a shortcut is often a circuitous detour.