The rapid advancement of artificial intelligence has led to a fierce competition among global technology giants, particularly in the fields of computational chips and foundational modelsAmidst this technological frenzy, a perceptive wave of capital is quietly positioning itself within the AI industry's "ultimate application layer"—the AI Agent sectorDescribed by OpenAI CEO Sam Altman as the "super interface of the AI era," this emerging domain has sparked structural changes in capital marketsAccording to statistics from consulting firm Pevc, since the beginning of 2024, the global AI Agent investment has eclipsed 66.5 billion RMBNotably, in the same year, AI startups in the United States garnered approximately 97 billion USD in venture capital, setting a new recordCompanies such as xAI, OpenAI, and Anthropic have attracted investments totaling billions, while Musk's xAI completed a 6 billion USD funding round in December 2024, with backers including prominent venture capital firm a16z.

With the dual propulsions of capital fervor and technological iterations, AI Agents are evolving from mere laboratory concepts into the pivotal engines of industrial transformationThis shift is being particularly pronounced in 2024, branded as the commercial year for AI Agents as major tech companies plunge into this space.

So, what exactly are AI Agents? In essence, they are intelligent entities capable of perceiving their environment, autonomously understanding contexts, making decisions, and executing actionsSimplified, an AI Agent operates as a sophisticated program rooted in large language models (LLMs) that can methodically accomplish assigned tasks through independent thinking and by utilizing various toolsBroadly, AI Agents can be categorized into autonomous agents and generative agents.

The distinction between AI Agents and LLMs lies in the interaction paradigmWhile LLMs rely heavily on human prompts, where the clarity and specificity of each prompt significantly influence the outcome, AI Agents primarily require a defined objective

Advertisements

They can engage in independent cognition concerning their goals, reflecting on past data and actions, learning from errors, and adjusting their future behaviors accordingly to adapt to their environmentsThis self-regulatory capacity makes them more effective in executing tasks and successfully achieving objectives.

In October 2024, buoyed by the capabilities of large models like ChatGPT, Microsoft unveiled ten autonomous AI Agents integrated into Dynamics 365, designed to automate customer service, sales, finance, and warehousing processesThese AI Agents, powered by OpenAI's o1 model and equipped with self-learning abilities, can execute extraordinarily complex, cross-platform business operations autonomouslyAn exemplary case is that of Lumen, a leading telecom company in the U.S., which estimates that its utilization of AI Agents benefits it by approximately 50 million USD annually, equating to the productivity of 187 full-time employees.

OpenAI introduced its first AI Agent, aptly named Operator, on January 24, 2025. This system was engineered to autonomously execute an array of intricate operations, which include coding, travel bookings, and automated e-commerce transactionsSubsequently, on February 2, 2025, the Deep Research feature was launched, aimed at deep research tasks, showcasing capabilities to compile professional reports in as little as 5 to 30 minutes, catering to high-intensity knowledge work across various fields, and relying on an o3 model for support.

Simultaneously, in January 2025, Anthropic rolled out a best practices guide for Agents, aimed at enhancing their efficiency and flexibility across multiple application scenariosA product called "AI Colleague," capable of writing and testing code, is on the docket for release in 2025. Its flagship model, Claude 3.5 Sonnet, demonstrated top-tier computer usability scores compared to other AI models during OSWorld testing.

Concurrently, in China, the launch of Alibaba Cloud's Tongyi Qianwen, which happened on January 29, 2025, marked the debut of its massive-scale MoE model-Qwen2.5-Max

Advertisements

With pre-trained data exceeding 20 trillion tokens, it excelled in various benchmark tests and outperformed DeepSeek V3 overallAlibaba’s Linguan Qwen also introduced a wholly open-source visual model, Qwen2.5-VL, available in sizes of 3B, 7B, and 72BNotably, both these models exhibited substantial improvements in performance, particularly in AI Agents' application across computer use—performing complex tasks like weather inquiries, flight bookings, and messaging autonomously.

As the Spring Festival approached, the entry of DeepSeek into the AI domain accelerated the ongoing commercial transformation brought about by technologyReports indicate that the DeepSeek-R1 employed large-scale reinforcement learning techniques during its post-training phase, achieving marked enhancements in reasoning capabilities, even with scant labeled dataThis technological leap positioned DeepSeek-R1's performance in mathematics, coding, and natural language reasoning on par with OpenAI's latest offerings, further realizing the vision of being open-sourceSuch developments signify a rapid move towards reducing costs while enhancing performance in the AI sector.

The continuous evolution of large models propels swift growth in the AI Agent industryThe iterative improvement of LLM algorithms feeds further capabilities into the AI Agents, making them essential on the pathway to achieving Artificial General Intelligence (AGI). Typically, the foundational architecture of an LLM Agent comprises five components: LLM, perception, planning, memory, and action, with LLM serving as the "brain" that enables reasoning and planningAs the costs associated with the reasoning endpoints of large models keep diminishing, AI Agents are set to flourish furtherAccording to Root Analysis, the global AI Agent market is forecasted to swell from 5.29 billion USD in 2024 to approximately 216.8 billion USD by 2035, boasting an impressive compound annual growth rate of 40.15% during this period.

Presently, AI Agents are being employed across various scenarios, including customer service, programming, content creation, knowledge acquisition, finance, mobile assistance, and industrial manufacturing, signifying a landmark shift from simple rule-based tasks to advanced autonomous intelligence

Advertisements

This evolution enhances productivity and transforms operational modalities, opening new frontiers in how humanity interprets and modifies the world.

With ongoing advancements in AI technology, these agents now transcend mere task execution; they exhibit the capability to operate independently based on complex logic, thereby facilitating efficiency and enabling breakthrough innovationsObservations from various sectors indicate that 2025 is set to emerge as the year for AI Agents' commercial applications, indicating that their adoption will happen at a pace faster than anticipated, heralding a significant surge in AI Agent technologies.

In tandem with these developments, the SaaS sector is experiencing a renaissance; as AI Agents respond to enterprises' needs to accelerate their intelligenceThe ongoing digital transformation within enterprises is deeply influenced by these agents, prompting SaaS platforms to evolve from basic management tools to being pivotal engines driving intelligent business operations.

According to a report by Jiemian News on November 21, 2024, NVIDIA CEO Jensen Huang, in conversations with notable Silicon Valley venture capitalists, stated that modern computing is transitioning from traditional data centers to "AI factories." These factories are not merely about storing and processing data; they are burgeoning centers for generating AI and agents, which are anticipated to become integral components of societal infrastructure across various industriesHuang emphasized that SaaS platforms will not be disrupted by AI, but rather they stand to be fertile grounds for nurturing innovations in agents. “They (SaaS businesses) are sitting on a goldmine,” Huang remarked, predicting the emergence of millions of AI agents that can enable more efficient intelligent management for specific tasks.

Northern America's renowned venture capital firm, Y Combinator, has also released commentary suggesting that with the accelerating improvements in AI models and their ensuing competition, a new commercial model is taking shape—vertical AI Agents.

Specific applications of AI Agents across various domains reveal immense commercial potential and prospects

Advertisements

Advertisements