Frost & Sullivan releases '2026 China General Intelligent Agent Enterprise-Level Implementation and Excellent Practices – Baidu Dazi Industry Examples Selection'.

A general intelligent agent is an intelligent system that possesses multimodal interaction and execution, cross-scenario and cross-domain adaptability, chain reasoning and autonomous learning, goal-oriented and delivery capabilities. Compared to dedicated intelligent agents, it does not rely on fixed templates and can serve as a basic capability component reused across multiple industries, offering broad coverage and fast deployment; compared to dialogue assistants and Copilot products, it moves from passive response and auxiliary suggestions to autonomous planning, delivering complete outcomes based on user goals and constraints.

Looking at the development history, 2023–2024 was the exploration period: the product was primarily dialogue plus plugins, with tasks being single-turn and single-tool; 2025 was the concept validation period: the deliverables changed from text to results, but the reliability of long-term tasks remained insufficient; 2026 is the explosive period: the focus of general intelligent agents shifts from individual to enterprise-scale deployment and begins to be bundled with office suites, operating systems, and smart hardware, with the core competition shifting to their ability to deliver real-world tasks.

Core capabilities of general intelligent agents and comparison with dedicated intelligent agents

The enterprise side has become the core growth driver for the industrialization of general intelligent agents. Currently, over 60% of enterprises have tried general intelligent agent products, and the mainstream business model is seat-based payment. The internet and technology industries have a strong digital foundation and ample data accumulation, resulting in the highest application maturity; the financial industry has significant technology investment, dense data, and clear compliance systems, making it a sector where value realization occurs quickly; industries such as industrial manufacturing, government services, retail e-commerce, scientific research and education, and energy and power are also accelerating their penetration.

However, an enterprise’s production environment is far more complex than personal use. The report summarizes the five major challenges of enterprise-level implementation: unstable delivery, requiring extensive rework; fragmented systems, making it difficult to integrate data and processes; lack of enterprise context, preventing experience from being captured; general lack of expertise and absence of specialized suites; data security and compliance concerns. Correspondingly, enterprise-level general intelligent agents must possess five capabilities: task execution and stable delivery, seamless integration with enterprise systems, team collaboration and knowledge retention, industry specialization and security governance.

In terms of implementation path, enterprises typically go through four stages: scenario pilot, department promotion, process embedding, and organization-level scale-up: first, validate value in high-frequency, clearly defined, and result-verifiable scenarios; then, convert personal experience into team standard capabilities; subsequently, embed it into formal business processes; and finally, make it an enterprise infrastructure. The key to this process is robust underlying intelligence of the intelligent agent and continuously controllable costs.

Challenges in enterprise-level implementation of general intelligent agents

Looking ahead, the report believes general intelligent agents on the enterprise side will exhibit four trends: first, manufacturers are accelerating the construction of enterprise context infrastructure to address the shortcoming of “not understanding the enterprise’s business”; second, the FDE deep delivery model is being deepened, allowing customers to jointly define business issues and solutions and complete the “last mile” value delivery; third, scalability is enhanced through the Skill mechanism and specialized suites, and the delivery format is gradually becoming more refined; fourth, credible security governance is upgraded to an organization-level infrastructure, enabling enterprise-level general intelligent agents to become identifiable, authorized, auditable, and traceable.

On the implementation level, the report recommends that enterprises advance from three aspects: in product selection, focus on delivery, connection, retention capabilities, and security—namely, the ability to directly produce verifiable finished products, the ability to connect with enterprise documents, OA, CRM, and other systems, the ability to solidify experience into reusable skills and knowledge bases, and the ability to support private or local processing; in scenario selection, prioritize high-frequency shared scenarios such as daily report writing and data retrieval, as well as data-intensive scenarios such as order aggregation and report generation; in permission control, prepare a comprehensive permission map in advance, gradually open permissions based on role-based minimum rights, and embed full-chain audit into the execution process.

Recommendations for the implementation of enterprise-level general intelligent agents

The report uses Baidu Dazi as a case study, including its practice cases in business services, manufacturing, online education, scientific research and education, smart hardware, and energy industries, covering scenarios such as recruitment and contract management, smart replenishment and smart office collaboration, course and student performance analysis, reuse of teaching and research experience, local offline AI workstations, grid operation ticket review and fault analysis. Among them, the business services case saw a 5-fold increase in resume preliminary screening capability, the online education case saw a reduction in task duration by approximately 90%, and the energy industry case saw an 92% increase in operational efficiency and a more than 12-fold increase in fault analysis efficiency. Baidu Dazi builds differentiation through professional depth and enterprise-level capabilities, and via the Harness engine, open connection, specialized suites, and multiple security protections, it supports stable delivery of enterprise-level tasks.

Baidu Dazi Industry Excellent Practice Case Map