Microsoft Unveils AI Agent Ecosystem with 1800 Language Models, Entering the Agent Era

On November 19th, local time, Microsoft announced a plethora of updates to its cloud and artificial intelligence services at the Ignite conference held in Chicago. The tech giant highlighted several enhancements to its Microsoft 365 Copilot platform. In addition, the company emphasized its growth in the AI sector, with Chief Communications Officer Frank Shaw stating that nearly 70% of Fortune 500 companies are now utilizing Microsoft 365 Copilot. Microsoft, like its large tech competitors, is investing billions of dollars to expand its AI capabilities, and Wall Street is eager to see returns on these investments.

The most noteworthy revelation at the conference was Microsoft's quiet establishment of a super AI agent ecosystem, the scale of which is hard to ignore. At Ignite, Microsoft disclosed the development status of the largest enterprise AI agent ecosystem. The company claimed that since its launch, over 100,000 organizations have created or edited AI agents through its Copilot Studio—a significant milestone that places Microsoft at the forefront of the market in this most watched and exciting enterprise technology domain.

Charles Lamanna, an executive at Microsoft responsible for the agent development vision, said in an interview, "The pace of agent ecosystem development is much faster than we imagined and faster than any other cutting-edge technology we have ever released. This business is doubling almost every quarter."

With Microsoft significantly expanding its agent business, rapid adoption by enterprise customers has followed. At Ignite, Microsoft announced that it would allow businesses to freely select from 1800 large language models (LLM) offered in the Azure catalog through these agents—a crucial move for Microsoft to reduce its full dependence on OpenAI models. The software giant also launched autonomous agents capable of operating independently, detecting events, and coordinating complex workflows with minimal human intervention.

Strictly speaking, an AI agent is software that relies on generative AI to reason and perform specific business tasks. These agents are now becoming powerful tools in the fields of enterprise automation and productivity enhancement. Microsoft's platform enables organizations to build agents for various use case scenarios, including customer service and complex business process automation, while maintaining enterprise-level security and governance standards.

Born for Enterprise Applications

Microsoft's early leading position in the AI agent field stems from its high attention to the practical needs of enterprises, which are often overlooked in the AI hype cycle. Although topics such as new autonomous agents and the flexibility of large language models have long dominated Ignite's attention, Microsoft's real advantage lies more in its solid enterprise infrastructure.

The platform integrates over 1400 enterprise systems and data sources, ranging from SAP to ServiceNow and SQL databases. This extensive connectivity allows organizations to easily build agents that can directly access and process data within their existing IT environments. In addition to choosing to build custom agents from scratch, Microsoft also offers ten pre-built autonomous agents specifically for core business functions such as sales, service, finance, and supply chain, accelerating the progress of common enterprise use cases.

Microsoft has not disclosed which type of agent is most popular among its customer base, but Lamanna stated that there are two types of applications: one for specific core tasks of the IT department and another that emphasizes a bottom-up approach. Employees can create Copilot agents to share documents or presentations with teams or other partners, allowing others to interact with the content and ask questions.

In the past, security and governance features were mostly considered after AI deployment, but now these two elements have been built into Microsoft's core architecture. The platform's control system ensures that agents always operate within the framework of enterprise permissions and data governance.

Lamanna explained specifically, "We believe that agents will be ubiquitous. Whenever there is a capability that makes the impossible possible, people will be amazed by its effects and eventually adopt it widely." He also compared AI agents to the internet, stating that the internet's coverage expanded from browsers to operating systems, fundamentally changing the client-server architecture.

In Lamanna's view, the significant breakthroughs in large language models have enabled software to understand unstructured content (including language, video, or audio) and perform preliminary logical reasoning, drawing conclusions or making judgments based on this data. "On the basis of this capability, browsers, word processors, core operating system experiences, and the implementation of sales processes and customer support will all undergo a major overhaul… In my opinion, as various agents and AI functions become widely available, every component in the computing stack will see some redesign."

Early adopters have also seen results. McKinsey used an automated routing agent to successfully reduce its project intake workflow from 20 days to just 2 days. Pets at Home deployed a fraud prevention agent in less than two weeks, saving millions of dollars annually. Microsoft also revealed that organizations using Copilot Studio include Nsure, McKinsey, Standard Bank, Thomson Reuters, Virgin Financial, Clifford Chance, and Zurich Insurance.

Agent Mesh:

Simplifying Collaboration Between AI Agents

The core of Microsoft's strategy is what Lamanna calls the "agent mesh"—a set of interconnected systems where AI agents collaborate to solve complex problems. Agents do not operate in isolation but can seamlessly pass tasks, messages, and knowledge within the entire enterprise.

So far, Copilot Studio has been associated with AI agents triggered by chat, but Microsoft is working to expand the range of operations. Imagine how enterprises will operate when agents collaborate seamlessly: a sales agent can request an inventory agent to check stock levels and then notify a customer service agent to provide updated information to customers. The features of this architecture include:

Autonomous agents that can detect events and trigger actions without human supervision;

Establishing a business process layer that coordinates multiple specialized agents;

Providing real-time monitoring tools for good transparency of agent workflows.

Microsoft's research department recently released the Magnetic-One system based on the company's Autogen framework. This system establishes a complex hierarchy of agents: management agents maintain task lists in the "outer loop," while specialized agents are responsible for execution in the "inner loop." This architecture is expected to soon adopt tools like Microsoft's OmniParser, with agents interpreting various UI elements, demonstrating Microsoft's leading technological reserves in operating computers through agents—also competing head-to-head with features being developed by Anthropic and Google. Microsoft stated that it is working to put this research into production but did not specify what form the product will take or when it will be released.

Microsoft Research's Magnetic-One multi-agent system emphasizes solving open Web and file-based operational tasks through inner and outer loops.

Microsoft's solution helps enterprise customers solve a key challenge: how to scale agents from hundreds to millions while maintaining control. The platform allows businesses to coordinate multiple specialized agents through its orchestration capabilities, a method that clearly aligns with the industry's reliance on and widespread adoption of multi-agent systems.

The platform's billing model also reflects this enterprise-first positioning philosophy. Unlike most AI providers who bill based on tokens, Microsoft Copilot Studio's pricing is based on the actual number of messages exchanged—emphasizing business outcomes rather than raw computing power. Lamanna explained that today's enterprise customers are no longer concerned about which model is best but demand use cases with real business value, "This represents a significant shift in the market."

Enterprise AI Agent Competition Intensifies

While other tech giants are also heavily investing in AI agents, Microsoft's combination of traditional enterprise functions and extensive integration gives it an early advantage. Competitors like Salesforce and ServiceNow have released their own AI agent platforms, such as Agentforce (claiming to have built tens of thousands of agents) and ServiceNow Agents. However, these products are relatively new and lack Microsoft's broad and solid enterprise coverage: for a long time, hundreds of millions of enterprise employees worldwide have been using the productivity suite created by Microsoft.

In addition, the strategic thinking of each competitor is different. OpenAI focuses on open APIs for direct access but has not yet established an enterprise AI agent deployment framework. However, its recently released o1 preview model demonstrates excellent reasoning capabilities, which are expected to support more intelligent AI agents in the future. Newcomers like Crew offer experimental agent frameworks but lack enterprise application scale. LangChain's modular framework remains popular among developers but focuses more on experimentation than enterprise deployment. Amazon Web Services maintains a developer-centric approach through platforms like SageMaker; Google's AI platform excels in specific vertical areas but lacks a unified agent framework suitable for widespread enterprise adoption.

In contrast, Microsoft combines enterprise security, low-code tools, pre-built templates, and professional code SDKs for developers, making it a more inclusive option for different teams. The software giant has also spent a lot of time integrating its various applications and analytics databases to ensure that AI agents can call on any enterprise data locally, eliminating the cumbersome process of单独 calling databases for retrieval-augmented generation (RAG) needs.

Of course, AI agent technology is still in its infancy. Large language models can still produce hallucinations, and it is necessary to maintain a cautious deployment and management attitude towards AI agents based on these models to avoid falling into infinite loops or causing unnecessary costs. Some customers have also expressed concerns about the challenges of Copilot in terms of pricing and practical application.

This part of the market may also be fragmented for a long time. A large part of the Fortune 500 companies tend to prefer a multi-vendor approach, such as using Microsoft's Copilot agents to improve employee efficiency while choosing other frameworks for sensitive application scenarios.

He who has AI Agents, has the world?

Although Microsoft is currently leading in enterprise AI agent deployment, the technology is still in its early stages of development. Microsoft's advantage does not lie in any single feature but in its comprehensive approach to achieving effectiveness: a high focus on enterprise infrastructure, extensive integration, and business outcomes themselves—not raw AI capabilities.

Whether Microsoft can maintain this leading position will undoubtedly depend on the market performance of its solutions in the coming year. Competitors are rushing to enhance their products, and enterprise customers are moving from the trial phase to full deployment. It can be imagined that AI agents are crossing the hype cycle and entering the practical application of enterprise IT architecture—and this transformation period will inevitably bring various complex factors and severe challenges.

For technology leaders, now is the best time to assess how AI agents can change established workflows. From automatically executing repetitive tasks to enabling new collaboration models, everyone should start small, focus on quantifiable results, and prioritize the use of pre-built agents to accelerate this journey of change. It is believed that whoever can take the lead in this wave of agent-driven business transformation will be able to establish an advantage in the next phase of market competition.