Representative End User Clients
Representative Automation Clients
Representative Software Clients
On October 1st, 2025, Microsoft officially unveiled its strategic vision for the next era of enterprise AI with the announcement of the Microsoft Agent Framework, marking a clear transition from assistive "copilots" to autonomous "agents." For years, we at ARC Advisory Group have tracked the convergence of IT, OT, and ET. Now, Microsoft is proposing a new, intelligent layer to sit atop this converged stack—an “open agentic web” where swarms of specialized AI agents collaborate to automate complex workflows. This moves the industrial paradigm from "data to insight to human action" to "data to insight to autonomous agent action."
This is a compelling vision for the future of the industrial enterprise. However, for our clients across manufacturing, energy, and supply chain, the path from today's plant floor to tomorrow's "agentic factory" requires a clear-eyed strategy. While Microsoft is laudably embracing open standards, its ecosystem creates a powerful gravitational pull toward Azure. The key to harnessing this new wave of AI is not to get swept away by the hype, but to anchor your strategy in a solid data foundation and a pragmatic, phased approach.
What Industrial Leaders Need to Know
Microsoft's latest announcement is not about a single product but a cohesive ecosystem. For our audience of IT, OT, ET, and Data Science professionals, here are the most salient takeaways:
The Dual-Engine Strategy: AutoGen and Azure AI Foundry: At the heart of the strategy are two complementary engines. First is AutoGen, an open-source framework for prototyping and research that allows developers to build and experiment with multi-agent conversational workflows. Second is the Azure AI Foundry Agent Service (https://azure.microsoft.com/en-us/products/ai-foundry/agent-service), a fully managed, enterprise-grade cloud service for deploying, scaling, and governing these agents in production. This dual approach provides a clear "prototype-to-production" pipeline, from open-source experimentation to secure, scalable deployment.
Enterprise-Ready Foundations: A major focus of the announcement was on making agents viable for mission-critical enterprise use. This is addressed through a suite of governance and security tools. Microsoft Entra Agent ID (https://azure.microsoft.com/en-us/blog/all-the-azure-news-you-dont-want-to-miss-from-microsoft-build-2025/) gives each agent a unique, manageable identity, allowing IT to apply familiar access controls. This is coupled with integrations with Microsoft Purview for data compliance and built-in observability dashboards for monitoring agent performance, cost, and safety, creating a much-needed governance wrapper for AI workloads.
The Critical Role of the Industrial Partner Ecosystem: Microsoft is not going it alone. The company is leaning heavily on its deep partnerships with industrial giants to bring this agentic vision to the factory floor. Siemens is co-developing its Industrial Copilot and bringing its Teamcenter X PLM software to Azure. Rockwell Automation is integrating its FactoryTalk® suite with Azure IoT and embedding a Copilot into its design software. PTC is developing a multi-agent model in Fabric to extend the digital thread, and AVEVA is enabling its AVEVA Connect platform to interoperate with Microsoft Fabric to unify operational data. These collaborations are essential for translating cloud AI capabilities into tangible OT outcomes.
The "Openness Paradox": Cautionary Credit is Due
Microsoft, like AWS and Google before them, deserves credit for its genuine commitment to fostering an open, interoperable agent ecosystem. The company has thrown its full support behind two crucial emerging standards: the Model Context Protocol (MCP) for agent-to-tool communication and the Agent-to-Agent (A2A) protocol for inter-agent collaboration. By building support for these protocols directly into its platforms, Microsoft is helping to lay the groundwork for a future where agents from different vendors can communicate and collaborate seamlessly.
However, this embrace of openness at the protocol layer masks a powerful lock-in dynamic at the platform layer. The full value of the Azure AI Foundry is unlocked through its deep, proprietary integrations with the broader Microsoft cloud. Microsoft Fabric is positioned as the central data foundation, and Microsoft Entra ID as the identity fabric for both humans and agents. Once an organization’s data and identity governance are deeply entwined with these services, migrating to a competing agent platform becomes a monumental undertaking.
Critically, for industrial clients, Microsoft Fabric is not an out-of-the-box Industrial Data Fabric. The hard work of contextualizing raw OT data—enriching sensor readings with metadata from MES, ERP, and PLM systems—is essential for any meaningful AI application. This "context engineering" must happen at the edge to avoid creating a data swamp in the cloud. Microsoft is heavily reliant on its industrial partner ecosystem to deliver this crucial capability, a significant dependency for clients to manage, especially as many of these partners maintain multi-cloud strategies that include AWS.
As we've covered at ARC, the industrial cloud landscape is now a fierce two-horse race. AWS has made significant progress in attracting the very same industrial software partners that Microsoft depends on, creating a vibrant multi-cloud ecosystem. While Azure often appeals to enterprises with its user-friendly interfaces and deep integration into the Microsoft stack, AWS is favored by teams that prioritize flexibility, developer control, and a more modular, framework-agnostic approach. This dynamic is playing out in the agentic space as well; AWS has its own robust multi-agent offerings like Amazon Bedrock Agents and the open-source Strands Agents SDK, which also embrace open protocols like MCP and A2A.
Paradoxically, this competition around open standards may ultimately benefit customers, creating a future where agentic solutions are more portable across clouds, giving industrial enterprises unprecedented choice and mitigating the risk of deep vendor lock-in.
Pragmatic Recommendations for Industrial Enterprises
So, how should industrial leaders navigate this promising but complex landscape?
Data Foundation First, Agents Second: The single most important prerequisite for success with Industrial AI is a robust data foundation. Before you invest heavily in agentic AI, you must prioritize the modernization of your Industrial Data Fabric. This means standardizing data acquisition from plant-floor assets with OPC UA and structuring that data into a Unified Namespace (UNS) architecture that provides a single source of truth for your operations. An AI agent is only as good as the data it can access; without a clean, contextualized data layer, any agentic initiative is doomed to fail.
Adopt a Phased, Pragmatic Approach: A "big bang" rollout of multi-agent AI is a recipe for failure. Instead, build internal expertise and prove value incrementally.
Start with Open Source: Use Microsoft’s open-source AutoGen framework to experiment with multi-agent concepts on a high-value, low-risk use case, such as automating a complex production report. This allows your team to build skills without a major upfront investment.
Pilot for Production: Once a use case is validated, pilot it on the enterprise-grade Azure AI Foundry Agent Service (https://azure.microsoft.com/en-us/products/ai-foundry/agent-service). The goal here is to rigorously evaluate the platform's security, observability, scalability, and total cost of ownership before making a broader strategic commitment.
Architect for Interoperability with Open Protocols: While the pull of the Azure ecosystem is strong, you can take architectural steps to maintain flexibility and mitigate lock-in. The most powerful strategy is to build your agentic layer on a foundation of open standards.
Mandate both MCP and A2A: Establish an enterprise-wide policy that your systems will communicate using both the Model Context Protocol (MCP) and the Agent-to-Agent (A2A) protocol. MCP standardizes how agents connect to tools and data sources (e.g., an ERP, MES, or database), creating a portable "tool" layer. A2A standardizes how agents discover and collaborate with each other, creating a portable "collaboration" layer.
By mandating both, you create a dual-layered abstraction that decouples your business logic from any single vendor's orchestration platform.
This discipline is not theoretical; major enterprise vendors like SAP, Oracle, and Salesforce, along with global system integrators like Accenture and Deloitte, are rapidly adopting these protocols, signaling a clear industry shift toward an interoperable agentic future.
This momentum from enterprise software and system integrators is a powerful signal. For the vision of the agentic factory to be fully realized, it will be crucial for the ecosystem of industrial innovators—the very companies building the Industrial Data Fabrics that ARC has covered extensively—to also embrace these open standards. This will ensure that agents can seamlessly interact not just with enterprise systems, but with the rich, contextualized OT data that platforms from specialists like Cognite and SymphonyAI provide.
The Final Assembly: Why Partners Will Build the Agentic Factory
Microsoft's vision for the agentic factory is compelling, and the foundational cloud and AI pieces are being assembled today. However, we must be clear: Microsoft is building the factory, but it will be its industrial partner ecosystem that equips it with the machinery.
The true realization of this vision—the specialized agents that optimize production lines, predict asset failures, and streamline supply chains—will depend on the deep domain expertise and intellectual property of partners like ABB, AVEVA, Emerson, Honeywell, PTC, Rockwell Automation, Schneider Electric, Siemens, Yokogawa, and the lively ecosystem of “guerrilla innovators” modernizing industrial systems engineering in the new era of AI. It is their mission, should they choose to accept it, to build upon this new framework.
For industrial enterprises, the path forward remains the same: build your data foundation first, experiment pragmatically, and design for openness. By doing so, you will be ready to integrate the powerful, partner-driven solutions that will truly define the intelligent, autonomous, and efficient industrial enterprise of the future.
Engage with ARC Advisory Group
For ARC Advisory Group recommendations for Navigating the AI Wars, Closing the Digital Divide by Embracing Industrial AI, assembling your Industrial-Grade Data Fabric, and governing and guiding major decisions about enterprise, cloud, industrial edge, and AI software, please contact Colin Masson at [email protected] or set up a meeting with me, or my fellow Analysts at ARC Advisory Group to find out more about our Executive Insights Service for Industrial organizations, and Industrial AI Insights Service for Vendors.