Chairman of Sinovation Ventures and CEO of 01.AI Kai-Fu Lee
When AI Truly Understands the World
From Technical Evolution to Outcome-Driven Organizational Transformation
In its early days, robotics advanced slowly because machines relied on rote commands and often failed when confronted with unfamiliar situations. In recent years, however, breakthroughs in world models have equipped AI with physical intuition and common sense. By simulating the outcomes of actions in advance, AI now functions as the brain of a robot, enabling it to navigate real-world environments with human-like adaptability. This breakthrough paves the way for fully autonomous production and personal assistants to become a reality.
Until world models are perfected to handle all unknown scenarios, factories remain the most practical entry point for AI into the physical world. Unlike the unpredictability of open environments, a factory can be designed as a setting optimized for robotic operation. This means AI need not contend with every variable at once but can instead achieve mastery within controlled contexts, specializing in the production of specific components with precision.
Although world models remain central to AI, developing these foundational systems requires immense capital and computing power, much like the winner‑takes‑all dynamics of the semiconductor industry. I anticipate that, in the long run, only about six global giants will have the resources to compete. Without massive user scale and sustainable returns, most companies will struggle to keep investing.
This division of labor mirrors the PC and mobile eras. While chipmakers and OS incumbents continue to enjoy substantial margins, the application layer will capture the lion’s share of both profit and reach. Consequently, the future of AI will not lie in superficial API calls. Success will require deep integration, where models are fine‑tuned and technologies adapted to specific scenarios, fusing applications with models at a fundamental level.
Another important AI trend is that the smartphone will not be the ultimate form of AI‑first hardware. The current phone architecture, which requires unlocks, app launches, and taps, creates seconds of delay that break the natural rhythm of human communication. Therefore, I believe the next generation of AI devices must include five defining attributes:
1 Voice-Driven: Voice is the most natural user interface, and the technology is now precise enough for everyday use.
2 Always-On: No unlocking or launching required.
3 Ambient Listening: Continuous environmental awareness, without reliance on wake words.
4 Infinite Memory: A digital second brain that logs and synthesizes everything seen and heard.
5 Near-Invisible: Ultra-miniaturized wearables: brooches, glasses, or bracelets. AI will shift from a passive executor of commands into an active analyst and a personal assistant capable of reasoning through the nuances of daily life.
However, the deeper impact lies in the logic of enterprise operations. I foresee 2026 as the inaugural year for large‑scale adoption of multi‑agent AI in corporate settings. Once AI combines physical intuition with executional capability, its role within an organization will fundamentally change. AI will no longer be merely assistive software but an employee in its own right. Its value will be measured by the tangible output it generates rather than API calls or subscription fees. For instance, at 01.AI, 80% of code is now AI-written, with human-authored code accounting for only 20%. While this has increased developer productivity by several-fold, the true value is not in cost savings but in the dramatic acceleration of time‑to‑market to maintain competitive leadership.
Under this trend, the enterprise model for AI adoption will formally move beyond the single-agent era and enter a new epoch of multi-agent collaboration.
Historically, companies have assigned single AI agents to isolated, repetitive tasks such as routine administrative work or expense reporting. However, the future battleground will be an enterprise’s ability to build a dedicated silicon team optimized for its core business. Within a multi-agent architecture, specialized AIs with diverse expertise will communicate, reason together, and autonomously deconstruct the company's most complex cross-departmental operations.
This transition from an assistive tool to a virtual employee is, at its core, a restructuring of organizational interests. As AI agents inevitably replace certain human roles, they will unsettle the vested interests and job security of middle management, who may slow adoption out of fear of being replaced. To fully unlock AI’s comprehensive potential, top decision-makers must dismantle organizational barriers from the top down. In doing so, they can elevate the enterprise from relying merely on tool intelligence to achieving higher‑order competitiveness through organizational intelligence and industry intelligence.
Amid this wave of transformation, the Hon Hai Research Institute has been entrusted with a core mission: to define the future. Its role must extend beyond technical support to pioneering the next generation of AI‑native hardware. Apart from developing miniaturized devices with five key attributes and achieving a seamless three-dimensional integration of software, hardware, and AI, the Institute must leverage the Group’s vast factories as the ultimate proving ground for Embodied AI. By developing world models that grasp physical laws and deploying pragmatic, high-efficiency AI agents such as wheeled chassis paired with robotic arms, the Institute will lead the push to carry AI from the digital into the physical world. This is not merely forward‑looking R&D; it is the vital strategic fulcrum that ensures Hon Hai can leap from contract manufacturing to defining product standards and mastering the industry’s core rules in the AI era.
