Kevin Kelly's WAIC Speech: In the next 5 years, keep a close eye on these 3 tracks

What is the ultimate destiny of AI? Will it be ever-smaller advanced large models, or robots that can enter every household? Will it replace humanity, or coexist symbiotically with us?

At the 2026 World Artificial Intelligence Conference (WAIC), Kevin Kelly offered his perspective on this question.

This tech visionary, who previously accurately predicted the rise of cloud computing, the Internet of Things, and virtual reality, has now set his sights on three even more distant frontiers:

• Humanoid robots that will take at least another decade to reach full maturity
• AI emotional capabilities that will completely reshape our existing perceptions
• A brand-new economic ecosystem built entirely around individual intelligent agents

The following is a carefully curated synthesis of Kevin Kelly's speeches and interviews during WAIC, a valuable resource worth saving and reading in depth for anyone concerned about the future of artificial intelligence.

Today, I will focus my discussion on the future development trajectory and potential untapped opportunities in the AI industry.

First and foremost, I want to emphasize that there are still a vast number of core unknowns and uncertainties in the field of artificial intelligence at present:

The ultimate evolutionary form of AI remains undecided, the industry landscape could move toward either centralization or decentralization, it is unclear whether artificial intelligence will cause mass displacement of human jobs, and the technological ecosystem could evolve in either an open or closed direction.

None of these critical questions will be resolved within the next one to two years, and many will still be in the exploration phase even five years from now. Yet it is precisely this uncertainty that harbors unprecedented opportunities for groundbreaking innovation.

The industry directions I will outline below, though full of unknowns, represent exceptional opportunities for everyone in this room to build entirely new industries and deliver disruptive, world-changing outcomes.

Over the next five years, the AI industry will see the emergence of three core frontier tracks brimming with both opportunities and challenges: namely, the agent economy, humanoid robots, and AI emotional capabilities.

Personal Intelligent Agents and the Agent Economy: Your "Extended Self"

The first core frontier is the construction of personal intelligent agents and the broader agent economic ecosystem.

In the future, every person will have their own exclusive, always-online personal intelligent agent, which can be embedded in all kinds of devices such as smart glasses and handheld terminals, accompanying the user at all times and deeply perceiving every one of their behaviors and needs.

Over time, these agents will understand you far better than you understand yourself. I define this phenomenon as the "extended self" — an entity that exists independently of your physical self, yet coexists symbiotically with you, serving as an extension of human capabilities.

In the future, massive numbers of interconnected personal intelligent agents will form a brand-new, fully connected agent economic ecosystem, where agents can independently coordinate tasks, complete transactions, and build their own shared credit systems.

The arrival of the agent era has also spawned a whole new set of industry questions and enormous opportunities. We need to redefine the ownership rights of intelligent agents: do agents belong to the companies that developed them, or to the end users who operate them?

At the same time, building a cross-agent trust system is a completely unexplored technical field at present. How to achieve secure interaction and trusted collaboration between different agents is a core technical challenge that must be overcome in the future, and it also contains massive innovation opportunities.

Humanoid Robots: The Most Complex Man-Made Creation of Humanity

Humanoid robots will be the most complex man-made products humanity has ever developed, far exceeding the complexity of all existing technological products we have ever created. The successful deployment and maturity of this product will require the collaborative efforts of countless top global talents working together to overcome technical hurdles.

It is not only equipped with the most sophisticated artificial intelligence systems, but also integrates precision hardware structures and a dedicated energy supply system, making the highly integrated combination of these three components an extremely challenging technical feat.

As a result, a fully mature humanoid robot that can truly enter ordinary households and adapt to all kinds of daily scenarios will represent one of the most difficult engineering challenges in the history of human technology.

Our current technology is only in its very early stages. The mainstream large language models of today have obvious shortcomings: they lack autonomous learning capabilities and long-term memory mechanisms, two elements that are the core foundation of human-machine collaboration and intelligent iteration, and are also the key to future technological breakthroughs.

We are not looking to phase out large language models — these technologies still have extremely strong practical value. But text-only cognitive intelligence alone is far from sufficient, and the industry needs an entirely new cognitive system to fill in the gaps of existing large language models.

Most existing AI models are trained on text-based knowledge, giving them only book-learned wisdom, but no real-world situational awareness.

The core direction for the future is to develop spatial intelligence and world models that can adapt to the three-dimensional physical world and perceive the laws of physics.

This new generation of AI will no longer learn purely from text data, but will deeply study the physical, chemical, and biological laws of the real world, accurately understanding the logic of real-world scenarios such as object balance, fluid motion, and spatial structures.

A large number of laboratories around the world, including many in China, are currently working to advance this frontier field. Spatial intelligence and embedded real-world models will become the core underlying capabilities of the next generation of AI, and an essential requirement for humanoid robots.

The successful deployment of humanoid robots relies on overcoming three core technical challenges:

First, spatial perception intelligence, which breaks free from the limitations of purely language-based cognition to adapt to three-dimensional real-world scenarios;

Second, the development of bionic robotic hands. The human hand is an extremely precise organ with integrated pressure and temperature sensing capabilities that far outperform ordinary mechanical clamps. The engineering development of high-precision bionic flexible robotic hands is enormously difficult, requiring contributions from large numbers of engineers and advanced AI technology;

Third, ultra-low-power energy technology. The human brain, as a supercomputing platform, consumes only 25 watts of power, and the entire human body uses only around 300 watts, while the energy efficiency of current humanoid robots falls far short of this human benchmark.

At the same time, humanoid robots designed to serve people need to achieve an extremely high reliability rating of 99.999%. Even the most highly automated robot factories in the world today only achieve a 99% task automation accuracy rate. Every incremental improvement in precision by an order of magnitude requires overcoming massive technical barriers.

At this stage, specialized industrial and agricultural robot technologies are already relatively mature. Precision agricultural robots, for example, can deliver targeted water and fertilizer to each individual crop to achieve refined farming, with deployment difficulties far lower than those of general-purpose humanoid robots.

However, fields like self-driving cars, which can also be considered a type of robot, have still not achieved full unmanned operation. They can run autonomously in 99.9% of scenarios, but the remaining 0.1% of extreme edge cases still require human intervention.

Overcoming that final 0.1% of challenges will require an investment of capital, technology, and human resources equal to the total effort required to solve the previous 99.9% of problems.

This clearly shows that the full maturity of general-purpose humanoid robots will require at least another ten years of dedicated research and development work.

AI Emotional Capabilities: The Most Disruptive Breakthrough for Public Perception

What role exactly will artificial intelligence play in human society in the future?

If I were to talk about the prospects of artificial intelligence over the next 10 to 15 years, that would be a much more realistic and reasonable timeframe. I believe that over this period, artificial intelligence will evolve to act more like a partner, a coach, a guide, and a co-pilot for people.

The current relationship model we see is "human + artificial intelligence": it is neither AI working entirely on its own, nor humans acting completely independently, but rather the two collaborating closely together.

This is exactly the direction we are moving toward in the future — whether solving problems or advancing work, everything will be carried out in the form of team collaboration.

Humans and artificial intelligence have two completely distinct modes of thinking: one is human cognition, the other is machine cognition. When these two thinking patterns are combined, their combined power will far exceed the output of either operating alone.

It is against this broader backdrop of human-machine collaboration that the integration and deployment of AI emotional capabilities will become the most disruptive breakthrough in artificial intelligence that reshapes public understanding.

Emotion is a human-scale capability that humans can perceive and intuitively understand without complex training, and it is also the core bond that enables deep human-AI connections.

Current AI systems can already accurately identify human emotions such as happiness, surprise, and fear through cameras, and generate appropriate corresponding responses.

For example, when it detects that a child is feeling down, it can proactively offer companionship and comfort — scenarios like this are already achievable today.

In the future, we will go even further, enabling AI to facilitate pet anthropomorphic communication and rich emotional interaction scenarios.

I firmly believe that AI and robots with complete emotional perception and empathy capabilities will be able to build genuine, profound emotional bonds with human beings.

The connection between humans and intelligent agents will not be a fake human-machine interaction, but a real, tangible emotional relationship — and this will become the core competitive advantage of future AI products.

AI is like a man-made extraterrestrial life form: even though it can replicate most human traits, it can never possess certain unique qualities of human nature. AI will not replace humanity, but the era of human-AI symbiosis has already arrived.

Closing Remarks

Breakthroughs in humanoid robot technology, the integration of AI with emotional intelligence, and the construction of a full-coverage intelligent agent ecosystem represent the biggest sources of uncertainty in the AI industry over the next five years — and also its most central innovation opportunities.

In this process, we need to maintain an open mindset to experiment with new technologies that emerge around us, and we must be willing to try repeatedly, because these new technologies are still not perfect at this stage.

New technologies like artificial intelligence and virtual "mirror world" glasses are still far from ideal today.

If you try them right now, you will most likely not like them. They may fail to meet your needs, or simply not be worth the price tag, but you still need to keep trying them.

This is exactly the philosophy of "moving forward through failure". Because if we ignore, ban, suppress, or deny these new technologies, we will lose the ability to steer their development in a positive direction.

The only way to maximize the strengths of these technologies while minimizing their downsides is to actually use them. So when you hear about a new technology, give it a few more chances to prove itself.

Further Reading: AI and Employment, Industry, and Organizational Transformation

The development of artificial intelligence is reshaping human work and organizational structures in a profound, gradual manner. It is not only changing the way tasks are executed, but also redefining the very nature and value of work itself.

The arrival of AI is not a sudden revolution, but a continuous, long-term evolution.

1. The Diffusion Rhythm of AI: From Exploration to Full Integration

AI applications will not completely replace traditional systems in the short term, but will gradually permeate every layer of society in an incremental way. The next ten years will be a critical phase of mutual adaptation between humans and AI.

This transformation involves far more than just technology — it is also a revolution in organizational structure and corporate culture. To effectively introduce AI, companies must redesign their workflows, adjust their performance evaluation systems, and restructure their decision-making frameworks.

The work systems we used in the past were designed exclusively for human employees. But in future organizations, AI will become a formal member of the team, requiring dedicated space in both corporate culture and operational structures.

This kind of adjustment will not be easy, but just as electricity transformed production structures during the Industrial Revolution, AI will drive fundamental changes in management and collaboration methods.

In this transformation process, small and startup enterprises are often at the forefront of adoption. Their flexible scale and flat organizational structures allow them to quickly experiment and make adjustments. In comparison, large corporations, with their complex processes and multiple hierarchical layers, will absorb new AI technologies at a much slower pace.

But regardless of company size, the integration of AI is an irreversible trend that no business can avoid.

2. Two Forms of AI Application: Products and Capabilities

Within enterprises, AI applications can be divided into two categories. The first category is AI products: meaning the company directly produces or sells AI-based products and services. The second category is internal AI applications: where enterprises deploy AI in their own operations and decision-making processes to boost efficiency and innovation capabilities.

In reality, internal AI applications often develop much faster. Many companies do not directly produce AI products, but have already widely adopted AI tools in areas such as process management, customer service, market forecasting, and risk analysis. This invisible, behind-the-scenes AI is becoming a new source of productivity.

A similar situation occurred during the electricity revolution. When electricity was first invented, it did not immediately transform factory structures. But once factories began rearranging their layouts to accommodate distributed electric motors, the entire industrial system underwent a qualitative leap forward.

AI is currently at a very similar stage: it is not just bringing new products to market, but will also trigger a full reconstruction of organizational forms. When knowledge and intelligence are systematically embedded into every part of an enterprise, departmental boundaries will gradually blur, collaboration will become far more flexible, and the speed of innovation will increase dramatically.

3. Industry Impacts of AI: Knowledge-Intensive Sectors Are First to Be Affected

The first industries to be impacted by AI are those built around knowledge as their core asset. Fields including software development, marketing, pharmaceutical research and development, education, finance, and insurance have already become the frontlines of AI penetration.

These industries all share a common trait: they are information-dense, language-driven, and logically structured — exactly the areas where AI excels.

Take customer service as an example. The emergence of AI customer service has not caused massive job losses. On the contrary, AI systems take over a large volume of repetitive communication tasks, allowing human customer service representatives to focus on handling complex problems and maintaining high-value customer relationships.

The addition of AI allows enterprises to offer longer service hours and higher quality support, while also spawning entirely new job positions such as "AI Customer Service Supervisor" and "Interaction Experience Optimization Specialist". This trend is equally evident in industries like translation, content creation, educational tutoring, and financial analysis.

The role of AI is no longer about "replacing human labor", but about "augmenting human capabilities". It takes over individual tasks, while preserving the core value of the work itself. This "task replacement - work restructuring" logic is becoming the defining core feature of labor structures in the AI era.

4. The Rule of Thumb for Technological Transformation: Three Iterative Trials

When enterprises introduce AI into their operations, they often go through a pattern I call the "three trials" rule. The first attempt is usually very costly and has a high failure rate; the second attempt reduces costs but still delivers very limited results; it is only on the third attempt, when costs have fallen significantly, that the application finally becomes truly mature. This rule applies not only to enterprises, but also to individual users.

In the early stages of AI adoption, failure is almost an unavoidable step on the path to mastery. To truly harness the power of new technologies, you must go through repeated cycles of experimentation and learning.

As a result, "fail first, then succeed" is a mandatory lesson for everyone in the AI era.

5. The Restructuring of Work: Tasks Disappear, Jobs Remain

AI will not make jobs disappear, but it will make individual tasks disappear.

Every job is made up of a number of specific tasks. AI takes over the parts that have clear rules, are predictable, and can be easily repeated. But a complete, full job always includes elements of responsibility, judgment, creativity, and human interpersonal interaction — all capabilities that remain uniquely human.

Professions such as doctors, teachers, lawyers, designers, and programmers