The path to AGI: from models to systems in the 2026 AI landscape

At the 2026 World Artificial Intelligence Conference (WAIC), the focus of the large model and generative AI track shifted significantly. Out of 130 participating entities, 83 concentrated on AI agents or intelligent applications, while only 18 remained focused on basic large models. This trend reflects a broader industry movement from evaluating model intelligence to assessing the real value created by AI systems. The fundamental unit of AI competition is evolving from individual models to integrated systems. In 2026, the global AI industry is at a critical juncture, with discussions about AGI (Artificial General Intelligence) moving from theoretical speculation to strategic planning. Organizations such as OpenAI, Anthropic, and Google DeepMind view AGI as the ultimate goal of AI development. Liang Wenfeng, founder of DeepSeek, outlined a six-step roadmap for AGI, emphasizing the integration of language models, chain of thought, agents, continuous learning, self-iteration, and embodied intelligence. He stressed that each stage must build on the previous one to avoid wasted effort. The core elements of AI development include computing power, data, algorithms, talent, and organizational structure. As of June 2026, China’s intelligent computing power reached 2185 EFLOPS, but the competition is shifting from single-chip performance to system-level efficiency. Data usage in China for AI training and inference reached 199.48 exabytes in 2025, with inference data surpassing training data for the first time. Liu Liehong, director of the National Data Administration, highlighted the transition from managing data to using data effectively. Algorithms are also evolving from micro-innovations to fundamental breakthroughs. The industry consensus is moving toward multimodal world models that understand physical laws, enabling AI to handle long-range tasks and autonomously break down complex goals into subtasks. High-performance computing engineers are in high demand, with a supply-to-demand ratio of 0.15, and AI scientists earn an average of 132,000 yuan per month. Organizational structures are undergoing systemic changes as companies integrate AI research and development. In China, ByteDance merged the Feishu product team into Douyin, Alibaba integrated multiple agent product lines into Qianwen Office, and Tencent consolidated its R&D resources. Globally, Amazon and Google DeepMind also restructured their teams to focus on foundational model capabilities and AGI development. The AI industry is transitioning from 'acquiring intelligence' to 'organizational intelligence.' While models set the upper limit of AI capabilities, success in the industry requires a holistic approach that integrates computing power, data, algorithms, talent, and architecture. DeepSeek’s recruitment poster emphasizes that AGI is not just a distant goal but a long-term vision requiring sustained effort and strategic perseverance. In 2026, the rise of AI agents has become a clear trend, with models advancing from 20 to 70-80 points in capability. Agents now perform complex tasks through observation, thinking, and action loops, rather than simple question-and-answer interactions. However, the performance of agents is fundamentally limited by pre-trained large models. Professor Tang Jie from Tsinghua University stated that pre-training provides models with world knowledge and basic reasoning abilities, and more data, larger parameters, and more computation are essential for scaling models. Ant Group’s Ling & Ring 2.6 technical report highlights that the transformation of models into agents redefines the optimization goals of AI systems, with reasoning ability rooted in pre-training. Li Qiang, vice president of Tencent Cloud, compared large models to engines and the engineering chain to the process of turning them into functional agents, underscoring the importance of foundational model capabilities in driving AI innovation.