Article by Alibaba Cloud
The industry consensus is clear: 2025 marks the dawn of the AI Agent era. Moving beyond today’s chatbots and copilots, an AI Agent is perceived as an intelligent system that can independently set goals, make complex decisions, and execute multi-step tasks with human oversight. This shift from passive knowledge tools to proactive, action-oriented systems represents a major leap in agentic capability.
As McKinsey highlights, the next frontier for generative AI is the transition from knowledge-based tasks to action-based tasks performed by agents. Equipped with advanced reasoning, planning, and real-time strategic adaptation, AI agents are poised to operate in sophisticated environments to accomplish real-world tasks. It’s no surprise that the market for agentic AI is projected for explosive growth in the coming years.
The growth of AI Agent: From Experimentation to Growing Adoption
We are already witnessing this shift from a mere chatbot to agents taking actions. Enterprises are actively testing and deploying AI agents to streamline certain workflows—from automating meeting scheduling, analytics report generation and debugging code, to preparing marketing and sales campaigns, reviewing resumes and handling customer inquiries.
Some forecasts indicate that by 2026, agentic AI will start mature from isolated experiments to broader deployment among businesses. This acceleration will be fueled by significant advancements in agentic capabilities, including increased autonomy, context awareness, tool integration, multimodal capabilities, personalization, long-term memory, security and alignment.
Yet for wider adoption of agents to address business demands, there is still a lot of work to do – currently AI agents rely on the underlying large language models to perform tasks, and to reduce hallucination and increase accuracy in a specific domain, agent service providers need to work closely with enterprise clients to develop tailored, specialized agents that can solve the day-to-day business issues with high accuracy (consider deploying a healthcare agent in a clinic).
Hence, how to develop agents that enable flexible tool use to solve issues, real-time intervention control for alignment, and intelligent context management to increase accuracy, is paramount to drive the wider adoption of AI agents in enterprises.
From ModelScope to AgentScope: Managing a growing herd of AI agents
While we are moving into an agent era, we see another critical evolution within it: To unlock enterprises’ AI values, it is pivotal to shift from standalone model or agent use, to collaborative, multi-agent systems deeply woven into the fabric of business operations. It is only through this collaboration that enterprises can harness AI at a scale sufficient to discover profound new business value and generate transformative growth opportunities.
Today’s platforms, such as Alibaba’s ModelScope which provide models as a service, have been instrumental in democratizing access to a vast library of individual AI models and helping businesses tap into different AI capabilities. However, the limitations of a single, monolithic model or agent become starkly apparent when facing dynamic, complex business problems. While a single agent can automate a task, it usually fails to orchestrate a process to accomplish sophisticated goals.
This challenge is giving rise to Multi-Agent Systems (MAS). Imagine a complex business problem not being solved by one powerful AI agent, but by a coordinated team of specialized agents. An investment decision, for instance, would need to be supported by a herd of agents: first, a finance analytics agent that works on the research of the potential investee’s fundamentals, financial reports and valuation assessment; then, a report analytics agent that reviews and summarizes investment firms’ previous reports on this potential investee; and finally, a news agent that provides an overview of the company’s media coverage and brand reputation.
For enterprises to truly unlock the value of AI at scale, multi-agent cooperation is not just an option but the key tipping point. Individual agents are powerful tools, but only when they learn to be good team players, can they maximize their individual capabilities and generate tangible, compound value at scale.
Agent Framework to Drive Multi-agent Collaboration
To make multi-agent cooperation a reality, robust agent frameworks are essential. These frameworks provide a centralized platform for multi-agent collaboration, coordinating diverse models, tools, and data sources and orchestrating various agents on different pieces of software to execute complex, multi-step actions autonomously.
Leading this charge is Alibaba’s AgentScope, an open-source, developer-centric framework built specifically for creating controllable, production-ready multi-agent applications. It features services including agent construction and orchestration, agent deployment and execution, as well as visual development and monitoring of agents. It covers the full lifecycle of agent development, deployment, and monitoring, while addressing the need for orchestrating multiple, specialized AI agents to work seamlessly together.
Agent frameworks such as AgentScope are more than a toolkit. It is a foundational technology that directly addresses the core requirement for orchestrating multiple AI specialists to achieve a common business objective. By enabling seamless collaboration among different agents, they will catalyze the rise of sophisticated multi-agent systems and fuel the widespread adoption of agentic AI we anticipate in 2026 and beyond.
The future of enterprise AI should not be solitary; it should be collaborative and coordinative.
