Microsoft Build 2025: Revolutionizing Software with Agentic AI and Open Standards

The arrival of Microsoft Build 2025 marks a dramatic inflection point in the evolution of software development and the digital landscape more broadly: the era of AI agents is no longer a theory or a futuristic aspiration but an operational, rapidly scaling reality. Every major announcement from Build this year drives home the message that artificial intelligence—specifically, agentic AI—is not just another productivity booster, but an architectural shift on par with the advent of the personal computer, the Internet, and the cloud.

AI Goes Agentic: What Does This Mean?​

Agentic AI describes autonomous, context-aware software entities—agents—that can interpret goals, make decisions, interact with environments, and act on users’ behalf across complex digital workflows. Microsoft’s vision for the open agentic web imagines a future where such agents permeate the infrastructure of the Internet, operating on behalf of individuals, teams, and entire organizations.
This future is coming into focus faster than many expected. Microsoft reports that over 15 million developers are leveraging GitHub Copilot, while more than 230,000 organizations—including a staggering 90% of the Fortune 500—have used Copilot Studio to build customized AI agents and automations. These aren’t just abstract demos; enterprises such as Fujitsu and NTT Data are deploying Azure AI Foundry agents to optimize sales and develop proposals, while Stanford Health Care is already orchestrating agents to streamline complex, high-stakes workflows in oncology.

The New Developer Centricity: Tools for the Agentic Age​

Microsoft’s announcements at Build 2025 center on making agent building, deployment, and governance accessible to the broadest community of developers, from AI specialists to line-of-business technologists.

Transforming GitHub Copilot Into a True Agent Partner​

The headline feature is the transformation of GitHub Copilot. Formerly known best as an in-editor suggestion engine, Copilot is now positioning itself as an asynchronous, agentic partner: a model that not only aids with code completion but autonomously tackles tasks like code reviews, deployments, and troubleshooting, even when developers are offline. This shift is bolstered by a suite of new capabilities:

• Prompt management and lightweight evaluations: Teams can experiment and iterate on model performance and interaction patterns, directly within GitHub, without switching contexts.
• Enterprise controls: Enhanced management tools allow organizations to securely govern how models and agents are used.
• Open-source Copilot Chat integration: Core AI-driven capabilities are being merged into the open-source VS Code repository, reinforcing Microsoft’s commitment to open and collaborative software development.

Independent validation from industry trackers indicates GitHub’s userbase is indeed enormous—over 150 million developers as of early 2025—making such agentic capabilities instantly available at unprecedented scale.

Windows AI Foundry: A Unified AI Platform for Developers​

For Windows developers, the new Windows AI Foundry represents a significant leap forward. Unlike fragmented toolchains of the past, Foundry provides a single platform for training and inference, with APIs that cover both vision and language tasks. Standout features include:

• Local and cloud model management: Developers can manage open-source language models via Foundry Local or bring proprietary models for fine-tuning and deployment.
• Seamless integration across client and cloud: Solutions built for the Foundry can be deployed at both the endpoint and datacenter level, enabling more flexible application architectures.

In a critical move for openness, Microsoft is ensuring that other leading models—such as xAI’s Grok 3 and Grok 3 mini—are not only interoperable with Azure AI Foundry but hosted and billed directly by Microsoft. This “model marketplace” now comprises over 1,900 partner- and Microsoft-hosted options, providing broad choice and competitive innovation in the AI stack.

Azure AI Foundry: From Single Agents to Complex Systems​

Azure AI Foundry, Microsoft’s all-in-one AI application platform, continues to expand its toolset. Noteworthy new features include:

• Model Leaderboard and Model Router: Developers and enterprises can keep up with the best-performing AI models in real time for specific tasks, and automatically route requests to the optimal model for a given query.
• Agent Service General Availability: Enables orchestration of multiple specialized agents to collaboratively solve complex tasks, facilitated by unified SDKs, support for Agent-to-Agent (A2A) and Model Context Protocol (MCP), as well as integrated open-source libraries like Semantic Kernel and AutoGen.
• Observability Dashboards: New built-in tools provide transparency into agent performance, quality, cost, and safety, with granular traceability. This is vital for enterprises needing to monitor not just what agents do, but how and why they make decisions—a frequent demand from regulated industries.

Enabling the Secure, Governed Growth of AI Agents​

AI agents introduce new challenges in identity, security, and compliance. Blind spots, sometimes termed “agent sprawl”—where agents are created rapidly and without oversight—could pose significant risks, from data leaks to architectural chaos.

Microsoft Entra Agent ID: Robust Agent Identity Management​

Microsoft’s answer is Entra Agent ID, now in preview, which automatically assigns unique, trackable identities to every agent created in Copilot Studio or Azure AI Foundry. This not only deters the proliferation of unaccountable agents but also integrates with Microsoft Purview for end-to-end compliance, data governance, and security controls.
By tying agent identity into the broader Entra infrastructure (formerly Azure AD), Microsoft enables robust auditing, permission management, and rapid response if any agent’s behavior strays from policy—a move lauded by some third-party security analysts as “industry-leading” for federated agent oversight.

Copilot Tuning and Multi-Agent Orchestration​

Copilot Tuning empowers organizations to leverage their proprietary data and workflows to train models with a simple, low-code interface, thereby ensuring agents are both domain-specific and aligned with company policies. Multi-agent orchestration allows these agents to collaborate—fusing their specialized domain knowledge to solve multi-faceted, end-to-end business problems. Early enterprise adopters say this is a game-changer in fields like legal services and healthcare administration, where knowledge is siloed and workflows are highly nuanced.

Open Standards: Pathway to an Agentic Web​

One of Microsoft’s boldly stated ambitions at Build 2025 is to avoid proprietary lock-in as the agent ecosystem scales. To this end, the company is pushing for adoption of Model Context Protocol (MCP) as the backbone for interoperability between agents, frameworks, and platforms—spanning GitHub, Copilot Studio, Dynamics 365, Azure, Windows 11, and beyond.
Microsoft and GitHub have both joined the MCP Steering Committee to steward this open protocol, which is designed to:

• Unify identity and authorization: Updated specifications let agents and LLM-powered applications access data/services using existing user credentials—a marked improvement in both convenience and security.
• Standardize server and content discovery: A new MCP server registry service lets any organization create public or private, up-to-date databases for MCP server entries, promoting discoverability and transparency for agent endpoints.

Early reactions from standards groups and open-source communities have been cautiously optimistic, praising Microsoft’s willingness to open up core infrastructure-layer protocols. However, some analysts warn of the dangers if one or two vendors ultimately dominate implementation or deviations emerge in competing “open” standards.

NLWeb: HTML’s Successor for the Agentic Era?​

Microsoft’s introduction of NLWeb is arguably the most radical long-term move. Dubbed by some insiders as “HTML for the Agentic Web,” NLWeb provides a uniform interface for websites to expose their content and services in a way intelligible and actionable by large language models and their agentic applications.
What sets NLWeb apart:

• Conversational endpoints: Websites can offer direct, semantically-rich interfaces that LLMs and agents can use to interact with their content—moving beyond basic scraping or brittle APIs.
• MCP server compatibility: Every NLWeb endpoint acts as an MCP server, ensuring consistent discovery and secure integration by third-party agents.

Critically, this framework could lay the technical foundation for a web that is as browsable by AI as it is readable by humans, empowering new forms of automation, aggregation, and information retrieval.

AI for Science and Society: Microsoft Discovery​

Microsoft is also aiming at high-impact use cases with Microsoft Discovery, a new extensible platform intended to supercharge scientific research. The premise is that agentic AI can massively accelerate “discovery pipelines”—be it for new medicines, climate research, or advanced manufacturing—by handling literature reviews, experiment design, and even protocol generation.
Though Discovery is still early in its public rollout, Microsoft’s demonstrations at Build showcase agents capable of mining scientific literature, suggesting experiment modifications, and flagging promising research leads. In regulated domains like biotech and pharmaceuticals, where time-to-market is crucial, even modest improvements in workflow speed can translate to major societal gains.
Industry watchers, including voices from Nature and Science journals, have echoed the criticality of such platforms but emphasize that reproducibility, data privacy, and explainability will be the litmus test for broader adoption.

Critical Analysis: Strengths and Risks in Microsoft’s Vision​

Notable Advancements and Strengths​

• Developer-first approach: By integrating AI and agentic tools within platforms already embraced by millions of developers (GitHub, Windows, Azure), Microsoft is ensuring rapid uptake and practical impact—not just theoretical aspiration.
• Commitment to openness: Adopting MCP and open-sourcing critical components like Copilot Chat signals a strong stance against walled-garden ecosystems, countering past criticisms of Microsoft’s proprietary tendencies.
• Security and governance focus: Proactive identity management, observability tooling, and seamless compliance integration address some of the biggest barriers to enterprise AI adoption.
• Rich toolkit for customization: Low-code/no-code agent builders, deep model customization options, and the model ecosystem marketplace foster both innovation and inclusivity.
• Cross-industry collaboration: Partnerships and co-development with leaders in health, science, financial services, and government showcase Microsoft’s seriousness about domain-specific, high-value applications.

Areas of Concern and Caution​

• Ecosystem lock-in risks: Despite the overtures to openness, the tight coupling of identity (Entra) and model deployment (Azure/Windows) may create practical dependencies on the Microsoft stack. Small developers and enterprises may find it difficult to extricate themselves if they want to pursue alternative cloud or model vendors.
• Open standards governance: The success of MCP and NLWeb hinges on truly collaborative, independent stewardship. Historically, “open” standards have sometimes splintered or been co-opted, reducing their effectiveness and limiting interoperability. The tech community will need to vigilantly watchdog Microsoft’s role as both contributor and dominant vendor.
• Security and compliance scalability: As agents multiply, real-world observability and governance become exponentially harder—not just a technical but a process and policy challenge. Even the best dashboards and controls can be outpaced by scale, as seen in other tech booms (e.g., containers, cloud sprawl).
• Agent autonomy versus user control: The more agents are empowered to act without direct supervision, the greater the risk of unintended actions or even abuses, especially when handling sensitive data or operating within opaque, interconnected systems.
• Explainability and reproducibility in science: AI-powered “hypothesis generation” tools can accelerate research but also risk propagating errors, especially if the underlying models are insufficiently transparent or the training data is biased.

The Road Ahead: Opportunities and Questions for Windows and the Wider Web​

Microsoft’s Build 2025 announcements are rightly being called a redefining moment for the world of digital development—not only in tech-forward sectors but across every industry where automation, productivity, and security are paramount.
For the Windows ecosystem, the integration of AI Foundry both at the edge and in the cloud represents a renaissance for Windows as a developer platform, not just an enterprise operating system. For organizations, new agentic capabilities promise both quantum leaps in productivity and genuine competitive advantages, especially as the cost and complexity of building domain-specific AI solutions plummet.
But there are still critical questions looming:

• Will agentic interoperability—through open standards—actually materialize, or will cross-vendor balkanization set in?
• Can enterprises keep up with the governance demands, or will “agent sprawl” become the new shadow IT?
• As agents become more capable, how will responsibility and oversight evolve—especially in high-stakes contexts like healthcare, law, and scientific research?
• Most fundamentally, will these advances truly democratize AI development, or will they further concentrate power in the hands of platform vendors and AI giants?

Conclusion: The Agentic Web Is Here—Will It Deliver?​

For now, Microsoft’s blend of technical innovation, strategic openness, and focus on developer empowerment merits cautious optimism. The company is staking its leadership on a future where AI agents are as fundamental as browsers or operating systems. If it can succeed in building—and, crucially, governing—the open agentic web, it will have fundamentally reshaped the fabric of the Internet for the coming decades.
As always, the measure of this revolution will not simply be the scale or sophistication of the technology, but whose interests it serves, who controls its levers, and how well—if at all—it can be held accountable by the wider world. The age of AI agents has arrived; what happens next will depend on how openly, securely, and wisely it is steered.

---

Source: Microsoft Microsoft Build 2025: The age of AI agents and building the open agentic web - ศูนย์ข่าวสารประเทศไทย