On the afternoon of September 4, Bloomberg reported that DeepSeek, a startup based in Hangzhou, is developing an artificial intelligence model with more advanced agent capabilities, aiming to perform multi-step operations on behalf of users with minimal instructions. This system can also learn and improve based on previous actions. The report states that DeepSeek’s plans for the new agent model (previously unreported) reflect a larger shift in the tech industry. In recent months, OpenAI, Anthropic, and Microsoft have each launched their own agent software to simplify personal and professional tasks. Bloomberg noted that DeepSeek and most companies in the industry aim to build increasingly autonomous AI systems that can initiate and execute complex real-world operations with little human intervention. However, so far, AI Agents typically require a significant amount of human supervision. AI Agents, also known as AI agents, are seen as a key focus area following large models, possessing capabilities for autonomous decision-making, task decomposition, and cross-application collaboration. The global AI Agent market is expected to grow from $5.1 billion in 2024 to $47.1 billion by 2030, with a compound annual growth rate of 44.8%. At the same time, the penetration rate of AI Agents in both large enterprises and small to medium-sized enterprises will significantly increase. By 2028, the penetration rate in large enterprises is expected to reach 25%, while that in small to medium-sized enterprises is expected to reach 15%. Several major development trends for AI Agents include: Deepening enterprise applications: A report by Capgemini indicates that most organizations (82%) plan to integrate AI Agents by 2026, primarily for tasks such as email generation, coding, and data analysis. By 2028, at least 15% of daily work decisions will be made autonomously by Agentic AI. Active consumer applications: AI assistant applications remain mainstream, with consumer scenarios (such as virtual assistants) having a much higher penetration rate than enterprise scenarios. Multi-Agent collaboration: In the future, multiple AI Agents will be able to work together, forming more powerful productivity. For example, in medical diagnosis, the collaboration between imaging analysis Agents and clinical decision Agents will improve the accuracy and efficiency of diagnoses. However, the development of AI Agents also faces several bottlenecks, such as: Insufficient tool invocation and collaboration capabilities: AI Agents often need to invoke various external tools when performing tasks, but most current AI Agents have limitations in tool invocation, being able to call only specific tools, which restricts the application range and functional expansion of AI Agents. Difficulties in multi-modal data fusion: There are differences between different types of input data such as text, images, and audio, and existing conversion technologies can lead to the loss of significant details in this data. If cross-modal data is inconsistent, it can lead to biases in the AI’s understanding of the environment, affecting the accuracy of final decisions. Limitations in system scalability: As the number of Agents increases, the complexity of the system grows exponentially, posing severe challenges to system scalability. Simply increasing the number of Agents may not only fail to improve performance but could also lead to decreased system efficiency. Computational power and storage limitations: As the context window size increases to millions, hardware memory bandwidth becomes a bottleneck. The response speed of AI slows down when processing large volumes of documents. Training multi-modal data also requires substantial storage space and support for efficient distributed systems. The textual content references and cites AI search results and articles from the Securities China WeChat account titled “After Hours, DeepSeek, Major News Reported!” by Chen Ming. Disclaimer: Investment involves risks, and the content of this article does not constitute any investment or financial advice.