#AI horizons 25-05 – Microsoft Build 2025: Advancing the Agentic Web with AI Innovations

Executive Summary

At Microsoft Build 2025, the company showcased a powerful portfolio of AI innovations aimed at mainstreaming the era of the “agentic web.” From enabling any website to become chatbot-ready via NLWeb, to the general availability of Foundry Agent Services and the democratization of AI model tuning through Copilot Studio, Microsoft reinforced its strategic AI leadership. GitHub Copilot remains the world’s most widely used coding agent and has now evolved into a full-fledged developer agent, capable of autonomous workflows. Microsoft’s announcements promise not just productivity boosts but foundational shifts in how users, developers, and businesses interact with AI.

Key Points

• NLWeb: A new open project enabling conversational interfaces on any website with a few lines of code.
• Copilot Studio Enhancements: Agent orchestration, computer use abilities, integration with enterprise systems, and open model support.
• Custom Agents for M365: Pre-built healthcare, HR, and customer service agents integrated with Microsoft Copilot ecosystem.
• GitHub Copilot Agent: Evolves from code assistant to autonomous development agent and remains the most used coding tool in its class.
• Discovery Platform: AI research platform for scientists combining specialized agents with graph-based reasoning.
• Agentic Search: Upgrades to Bing and internal AI search layers for grounding and semantic querying.
• Democratized AI Tuning: Copilot Studio now lets business users tune LLMs with their own data, no ML expertise required.
• Foundry Agent Service GA: Microsoft’s hosted agent service moves to general availability, simplifying custom AI deployments.

In-Depth Analysis

NLWeb: Enabling Natural Language Interfaces on the Open Web

Microsoft’s NLWeb is a foundational technology for the agentic web, offering plug-and-play natural language interfaces for any website. By turning structured website data into AI-readable formats compatible with the Model Context Protocol (MCP), NLWeb allows conversational agents to interact with content live. Early adopters include TripAdvisor and Shopify, and the open nature of the initiative promises wide compatibility across AI ecosystems.

GitHub Copilot Agent: Leading the Charge in Autonomous Coding

Microsoft highlighted GitHub Copilot as the most adopted coding agent in the world, now capable of acting as a full developer agent. Beyond code completion, it can plan tasks, conduct reviews, fix bugs, and orchestrate deployments. Integration with Azure DevOps and GitHub Actions extends its reach into CI/CD pipelines, making it a true productivity multiplier for developers.

Copilot Studio: Tuning and Orchestrating Agents at Scale

Copilot Studio received major upgrades. Users can now:

• Tune AI Models with Business Data: Without requiring ML expertise, users can ground LLMs with their own structured content.
• Orchestrate Multiple Agents: New visual tools support agent collaboration on multi-step tasks.
• Access a Library of Pre-Built Agents: Microsoft released industry-specific agents (e.g., Healthcare, HR, Sales) ready for integration in Copilot for M365.
• Enable Computer Use: Agents can now interact with desktop applications and websites, a capability that vastly extends automation potential.

Industry-Specific Copilot Agents

New vertical-specific agents announced at Build include:

• Healthcare Agent: Automates chart summarization, patient history synthesis, and EMR interactions.
• HR Agent: Handles onboarding workflows, answers policy questions, and schedules interviews.
• Customer Support Agent: Automates triage, case classification, and contextual replies in integrated support platforms.
  These agents are deployable through Copilot Studio and Copilot for Microsoft 365.

Microsoft Discovery Platform: AI for Scientific Acceleration

Discovery combines AI agents with graph-based knowledge mapping to support scientific research. Integrated with Azure HPC and Azure AI Foundry, it allows contextual search, real-time hypothesis testing, and model routing. Microsoft researchers used Discovery to design a novel data center coolant in under 200 hours.

AI Search: Toward Agentic and Context-Aware Search Experiences

Microsoft is reimagining search around AI agents. As Bing Search APIs are sunset, new agentic search methods combine real-time grounding, model responses, and semantic ranking. “Grounding with Bing” allows AI agents to retrieve real-time information while maintaining context from user conversations.

Foundry Agents Service Goes GA

The Foundry Agents Service is now generally available, enabling enterprises to create, host, and manage custom agents using Azure AI infrastructure. It includes:

• A model router for optimizing LLM use
• Leaderboards for performance comparison
• Full support for retrieval-augmented generation (RAG)
  This lowers the barrier for enterprise-grade AI development while providing built-in monitoring and security.

Business Implications

Opportunities:

• AI-driven Productivity: Autonomous agents reduce cognitive load and accelerate workflows.
• Faster Innovation: NLWeb and Foundry agents reduce the time from idea to implementation.
• Domain-Specific Value: Pre-built vertical agents unlock sectoral efficiencies in healthcare, HR, and support.

Risks:

• Ecosystem Lock-In: Microsoft’s integrated AI stack could raise switching costs for businesses.
• Trust and Accuracy: AI hallucinations remain a risk; grounding and model validation are critical.
• Job Displacement: Autonomous agents may alter workforce compositions, especially in routine digital roles.

Market Impact:
Microsoft’s aggressive push across both developer and enterprise ecosystems signals a race to define the standards and platforms of the agentic web. With GitHub, Azure AI, and Copilot Studio, Microsoft is well positioned to own the tools, the infrastructure, and the interfaces of this next computing wave.

Why It Matters

Microsoft Build 2025 marked a turning point in mainstream AI adoption. From GitHub Copilot’s autonomy to the launch of ready-to-use agents in business apps, Microsoft has taken major steps to operationalize AI. The general availability of Foundry Agent Services and the ease of tuning LLMs in Copilot Studio place advanced AI capabilities into the hands of non-technical users. This democratization will shape competitive dynamics across industries, making AI integration not just an option, but a necessity for future-ready enterprises.