en.Wedoany.com Reported - On June 18, the China Academy of Information and Communications Technology (CAICT) released the Top 10 Keywords for Intelligent Agents in 2026, providing forward-looking insights into technological breakthroughs, application deployment, and ecosystem development for intelligent agents. The ten keywords are: Intelligent Agent Infrastructure, Intelligent Agent Interconnection and Collaboration, Intelligent Agent Engineering, Intelligent Agent Learning and Evolution, Intelligent Agent Memory, Intelligent Agent Skills, Intelligent Agent Product Innovation, Intelligent Agent Payment Protocol, Intelligent Agent Trustworthiness, and Intelligent Agent Full-Stack Evaluation.
Intelligent Agent Infrastructure is the foundational system supporting the development, deployment, operation, and governance of intelligent agents. It encompasses resource bases such as computing power, storage, networks, and cloud resources; engineering support including sandboxes, development frameworks, deployment platforms, and runtime environments; key components like model access, memory management, tool invocation, skill orchestration, task scheduling, and state management; as well as governance mechanisms including observability, evaluability, auditability, security isolation, and permission control. As intelligent agents transition from pilot exploration to large-scale application, handling complex tasks demands higher performance in high concurrency, long duration, multi-tool, and multi-permission operations. Industry stakeholders are accelerating the construction of integrated environments for development, deployment, and operations, driving infrastructure towards standardization, modularization, and cloud-edge collaboration. In the future, Intelligent Agent Infrastructure will become a crucial cornerstone supporting multi-agent collaboration, complex task execution, and full-process business deployment.
Intelligent Agent Interconnection and Collaboration refers to the capability system enabling different intelligent agents, models, tools, and business systems to achieve interconnection, intercommunication, and interoperability through standard interfaces, collaboration protocols, and task orchestration mechanisms. Its core lies in promoting the evolution of intelligent agents from standalone operation to group collaboration, allowing different agents to exchange information, divide labor, and collaborate on complex tasks, forming a reliable collaboration network underpinned by mechanisms such as identity management, permission control, and security auditing. Intelligent Agent Interconnection and Collaboration bridges the connection channels between virtual and physical agents, facilitates closed-loop collaboration in decision-making, perception, and execution, and promotes cross-system, cross-platform, and cross-organizational collaboration, fostering an open intelligent service ecosystem. In the future, it will further break down system barriers and application boundaries, supporting multi-agent collaboration, complex task processing, and the development of an intelligent interconnection of everything ecosystem, becoming a key foundation for collective intelligence and large-scale application of intelligent agents.
Intelligent Agent Engineering is a critical pathway for advancing intelligent agents from concept validation to large-scale application, aiming to address deployment challenges such as non-standard development processes, uncontrollable operations, and unguaranteed quality in complex scenarios. Due to the characteristics of intelligent agents, including open-ended goals, dynamic paths, and uncertain results, it is necessary to build a full lifecycle closed-loop system covering requirements analysis, architecture design, development and construction, testing and validation, deployment and operation, monitoring and optimization, and cost evaluation. The focus is on advancing asset and process engineering. On one hand, models, tools, skills, processes, and knowledge are consolidated into reusable assets. On the other hand, system stability and controllability are enhanced through standard processes, operational monitoring, effect evaluation, and continuous optimization, creating a business closed-loop that is "repeatable, observable, and optimizable." In the future, as engineering methods, operational controls, and evaluation feedback mechanisms continue to improve, Intelligent Agent Engineering will become the core engine supporting the large-scale and controllable application of intelligent agents.
Intelligent Agent Learning and Evolution refers to the process where intelligent agents continuously improve their performance and adaptability through ongoing interaction with the environment, users, and tasks, by means of experience reflection, memory reconstruction, strategy updates, and capability leaps. Its core lies in replacing static capability presets with dynamic closed-loop optimization, integrating stages such as perception and understanding, task planning, action execution, and result feedback. This enables agents to continuously absorb interaction experience, optimize task strategies, refine memory structures, and adjust behavioral patterns, thereby enhancing adaptability, stability, and task completion quality across different scenarios, achieving a leap from passive "instruction-driven" to active "self-growth." In the future, Intelligent Agent Learning and Evolution will form a technical paradigm of continuous feedback, dynamic optimization, and long-term gains, enabling systems to accumulate experience, correct deviations, and enhance capabilities in complex environments, providing crucial support for long-term operation and complex task processing.
Intelligent Agent Memory is a core capability supporting contextual continuity, user understanding, and experience accumulation, helping agents maintain continuity, stability, and long-term adaptability during interaction and task execution. Unlike static information storage in traditional systems, Intelligent Agent Memory places greater emphasis on the dynamic management of interaction history, task status, user preferences, environmental feedback, and execution experience. It must support both context tracking and intermediate process recording within a single task, as well as information consolidation and experience reuse across sessions and tasks. The key lies in achieving effective organization, accurate retrieval, and secure control of memory through mechanisms such as information storage, semantic retrieval, memory updates, content compression, active forgetting, and permission governance. In the future, Intelligent Agent Memory will evolve towards greater structure, schedulability, and adaptability, enabling agents to continuously enhance task understanding, strategy optimization, and long-term evolution capabilities while maintaining continuous experience and a unified identity.
Intelligent Agent Skills are key mechanisms that abstract and encapsulate specific operations, business rules, and domain knowledge involved in task execution into callable, composable, and reusable units, effectively compensating for the capability gaps of intelligent agents in diverse tasks within complex dynamic environments. Unlike relying solely on model-generated results, skills enable agents to invoke external capabilities such as retrieval, analysis, computation, generation, system operations, and device control around task objectives, and dynamically orchestrate them in conjunction with scenario constraints, execution feedback, and security rules. Through skill-based design, repetitive tasks can be standardized, business rules can be structured, industry knowledge can be toolified, and execution processes can incorporate verification, rollback, permission control, and security fault tolerance mechanisms, thereby improving the accuracy, stability, and transferability of task completion. In the future, Intelligent Agent Skills will evolve towards standardization, openness, cross-domain reuse, and autonomous orchestration, strongly supporting the transition of intelligent agents from single-task responses to full-process execution of complex business operations.
Intelligent Agent Product Innovation is a significant manifestation of the deep integration of intelligent agent technical capabilities with application scenarios, marking the transition of agent applications from functional validation to a new stage of scenario-based, productized, and service-oriented development. The new generation of intelligent agents is no longer limited to question-and-answer interactions but has preliminarily acquired capabilities for task understanding, process scheduling, tool invocation, and system operations. Through multi-terminal access, permission control, skill expansion, and memory consolidation, they can bridge the complete chain from user interaction to task execution, and from front-end entry points to back-end systems. From the surge in practical applications of agents like "shrimp farming" and "horse breeding" to the continuous emergence of product forms such as intelligent assistants, intelligent customer service, and digital employees, intelligent agents are expanding from single dialogue interfaces to diversified product systems covering office, production, daily life, and industry services. In the future, Intelligent Agent Product Innovation will continue to develop towards personalization, proactivity, scenario-specificity, and ecosystem orientation, further enhancing user demand understanding, autonomous task planning, and business system linkage capabilities, gradually becoming digital partners connecting users, scenarios, and services.
Intelligent Agent Payment Protocol is a new rule system designed for autonomous transactions, service invocation, and value exchange by intelligent agents. It can significantly lower the barriers and costs of automated payments while addressing issues in traditional payment systems when applied to agent scenarios, such as limited subject eligibility, ambiguous responsibility attribution, and insufficient adaptability to dynamic terms. The Intelligent Agent Payment Protocol features flexible rule configuration, transparent processes, verifiable results, and traceable responsibilities. Combined with commercial trust mechanisms, it provides standardized support covering identity recognition, permission authorization, service interaction, automatic settlement, and trust assurance for cross-platform, cross-terminal agent collaboration, driving the evolution of agents from information relay nodes to transaction execution entities. In the future, the Intelligent Agent Payment Protocol will further refine mechanisms for autonomous transactions, service invocation, and settlement by agents, promoting trusted value exchange across systems, platforms, and organizations, providing crucial support for transaction execution and service coordination in complex business and multi-agent collaboration scenarios.
Intelligent Agent Trustworthiness is a crucial prerequisite for ensuring the standardized deployment, stable operation, and sustained application of intelligent agents. Its core lies in equipping agents with capabilities for reliable generation, controllable execution, transparent decision-making, compliant interaction, and responsibility traceability in complex tasks and open environments. As agents move from information Q&A to task execution, they engage in longer business chains, invoke more external resources, and have more direct real-world impacts. Issues such as unstable knowledge sources, unclear behavioral boundaries, uncontrollable execution processes, opaque decision paths, unverifiable result quality, and untraceable responsibility chains become more prominent. Enhancing Intelligent Agent Trustworthiness requires coordinated efforts in three areas: data, technology, and evaluation. This involves consolidating professional reliability through high-quality data supply and knowledge governance; ensuring task execution controllability through trusted reasoning, permission control, behavioral constraints, and risk protection; and achieving problem identification, risk quantification, and continuous improvement through multi-dimensional evaluation frameworks and risk monitoring mechanisms. In the future, Intelligent Agent Trustworthiness will continue to strengthen reliability, controllability, robustness, and traceability, building a foundation of trust for the large-scale application of agents and supporting their high-quality, sustainable development in complex, open scenarios.
Intelligent Agent Full-Stack Evaluation is a systematic evaluation system targeting the technical capabilities, application value, and operational benefits of intelligent agents. It is an important tool for assessing whether agents can be truly deployed, operate continuously, and create value. Unlike model evaluation, agent evaluation needs to cover multiple aspects including task understanding, planning and reasoning, tool invocation, execution feedback, stability, security, and user experience. It helps developers identify capability boundaries and technical shortcomings, supports business leaders in assessing scenario fit, user value, and strategic alignment, and provides a basis for managers to conduct cost-benefit analysis and return-on-investment calculations. CAICT's "Fangsheng" Intelligent Agent Benchmark is continuously refining evaluation methods around core capabilities, general tasks, and industry scenarios, promoting the coordinated development of capability evaluation, value evaluation, and benefit evaluation. Capability evaluation focuses on task completion quality, complex task handling, and industry adaptation levels; value evaluation focuses on user experience, business contribution, and strategic alignment; and benefit evaluation focuses on task automation levels, efficiency improvements, cost optimization, and investment returns. In the future, Intelligent Agent Full-Stack Evaluation will focus on real business scenarios and dynamic task environments, forming quantifiable, reproducible, and implementable evaluation methods, providing a basis for quality improvement, value judgment, and risk management in agent applications.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
