At NRF 2026, Microsoft positioned agentic AI as the next architectural layer for retail e-commerce, moving beyond isolated copilots and pilots toward a more coordinated, protocol-driven operating model. Rather than emphasizing individual AI features, Microsoft focused on how AI agents can sit on top of ERP and commerce systems to orchestrate end-to-end retail journeys across channels.
The message? Competitive retailers will need to re-architect for agent cooperation, not simply add AI interfaces to existing workflows. Rather than signaling the arrival of another copilot, Microsoft’s NRF 2026 messaging points to a longer-term shift: ERP and commerce platforms evolving into protocol-driven foundations for agent-orchestrated retail operations.
From Point Tools to Agentic Stack
Microsoft framed its retail strategy around an agentic architecture built on Model Context Protocol (MCP), combined with its own Agent Communication Protocol (ACP). Together, these are intended to give AI agents governed access to shared business context, including product data, pricing, promotions, inventory, policies, customer intent, carts, orders, and fulfillment options.
Central to this vision is the previewed Dynamics 365 Commerce MCP Server, expected in early 2026. The server is designed to expose core commerce capabilities—catalog, pricing, promotions, inventory, carts, orders, and fulfillment—as MCP-enabled services that agents can invoke. The goal is to allow agents to discover, decide, and execute actions using consistent business logic across digital, store, and conversational channels.
This approach positions protocols, rather than custom integrations, as the connective tissue between ERP-grade logic and customer-facing experiences.
Embedded, Custom, Partner Agents
Instead of promoting a single retail copilot, Microsoft outlined a multi-layer agent model:
- Embedded agents within Dynamics 365 aligned to merchandising, supply, store operations, and service use cases.
- Custom agents built using Copilot Studio, allowing retailers to encode proprietary logic while relying on MCP and ACP for secure access to enterprise data.
- Partner agents from global system integrators and independent software vendors expected to package vertical-specific patterns for segments such as grocery, fashion, and large-format retail.
The emphasis is on agents acting as governed participants in operational workflows, not simply as user-facing assistants.
What This Means for ERP Insiders
Microsoft is positioning Dynamics 365 beyond a retail application suite toward an ERP-adjacent orchestration layer for e-commerce. By exposing inventory, pricing, orders, and fulfillment through agent-consumable services, Microsoft is tightening the link between back-office systems and front-of-house experiences.
More broadly, Microsoft is positioning itself as a player in the emerging protocol layer for enterprise AI—one that emphasizes permissions, compliance, and cross-system coordination, with retail as a primary showcase industry. For ERP and retail technology leaders, Microsoft’s direction highlights several implications:
- ERP logic must be agent-ready. Core data and rules that are not exposed through standard, governed interfaces will be harder to use safely in AI-driven workflows. ERP platforms that cannot present clean, permissioned business logic to agents risk falling behind in the intelligent operations race.
- Architectures must assume multiple cooperating agents. Embedded, custom, and partner agents introduce new requirements for observability, governance, and integration across ERP-centric environments. The strategic shift is from managing single copilots to designing agent ecosystems, where coordination, conflict resolution, and accountability matter as much as model quality.
- Ecosystems become execution layers. Independent software vendors and global system integrators gain a clearer surface area to build repeatable agent patterns, shifting differentiation from UI features to operational playbooks. For the market, this means competitive advantage will increasingly come from how quickly partners can industrialize agent-driven processes, not just demonstrate AI capabilities.
