AI is moving beyond copilots toward systems that can reason, make decisions, and take action. Microsoft’s 2026 Work Trend Index[1] reports a 15× year-over-year increase in active agents in Microsoft 365, while its 2025 research found that 46% of leaders say their organizations are already using agents to fully automate workstreams or business processes.[2] For organizations already running Dynamics 365 Finance & Operations, this shift presents an opportunity to take their existing ERP investment further by enabling more intelligent and autonomous ways of working.
Dynamics 365 Finance & Operations is well positioned for this shift, with Microsoft embedding AI across the platform and introducing the Dynamics 365 ERP Model Context Protocol (MCP) to connect agents with ERP data and business logic. For existing D365 F&O users, this opens new possibilities across finance, procurement, supply chain, and operations. This article covers agentic AI in D365 F&O, the role of MCP, the agents available today, building custom agents with Copilot Studio, and the governance needed to deploy them responsibly.
From Automation to Agents: What’s Really Different?
Before diving into D365-specific details, it is worth making sure we are all on the same page about what “agentic AI” actually means, because it is easy to confuse it with technologies that already existed.
Traditional automation follows fixed rules: if an invoice is received, route it to the approver. It executes predefined steps without adapting to context.
Copilot introduced natural-language assistance, helping users find information, generate content, and summarize data. But it remains largely reactive, responding when prompted.
Agentic AI takes a different approach. Instead of being given every step, an agent is given a goal. It can determine what needs to be done, make decisions within defined boundaries, and execute multi-step workflows with limited human intervention.
Simply put: automation follows rules, Copilot responds to requests, and agents work toward goals.
Why D365 F&O Is Positioned for This
One of the reasons Microsoft’s agentic AI push is particularly interesting for Dynamics 365 Finance & Operations customers is that the platform was already deeply integrated across the business. Finance, supply chain, manufacturing, sales, and HR all operate on shared data models inside D365. That is a major advantage when it comes to AI agents.
Agents need data to reason over, and the more current that context is, the better decisions they can make. Because D365 F&O spans the full operational lifecycle of a business, agents deployed within it do not have to work with snapshots or exports. They work with live transactional data, live financial signals, and real-time operational information.
Additionally, because D365 agents are built on the same security model as the platform itself, they respect role-based access control, data isolation, and existing approval workflows. This is not a third-party tool bolted on from outside. It is intelligence embedded directly into your operational system.
The Engine Underneath: Model Context Protocol (MCP)
MCP is an open standard developed to solve a specific problem: how do you give an AI agent reliable, structured access to the data and business logic inside an enterprise system, without writing a custom API integration for every single scenario?
Microsoft’s answer was the Dynamics 365 ERP MCP Server, a dynamic framework that sits between agent platforms (like Microsoft Copilot Studio) and the F&O environment. Through this server, an agent can:
- Read and write data: create, update, and query records across D365 entities
- Execute business logic: trigger workflows, run business rules, and invoke actions the same way a user would through the application interface
- Access analytical data: query live financial metrics and operational signals for reasoning
An earlier version of the MCP server had a fixed set of 13 predefined tools, useful but limited. The new dynamic server adapts as your business evolves, letting you configure which tools and capabilities are exposed. Microsoft has announced the older static server will be retired on October 1, 2026, so if you are still on the legacy setup, now is a good time to migrate.[3]
What Agents Are Available Right Now?
You do not have to build everything from scratch. Microsoft has already released a set of purpose-built agents designed for common finance and operations scenarios:
Account Reconciliation Agent. Automates matching and clearing transactions between subledgers and the general ledger, a task that consumes enormous amounts of finance team time, especially at month-end. It flags discrepancies for human review rather than replacing judgment on exceptions.
Supplier Communications Agent. Takes over routine supplier interactions, drafting and sending accurate, timely responses using purchase orders, invoices, and delivery status, escalating to human buyers only when judgment is genuinely required.
Time and Expense Agent (Project Operations). Automates entry and approval so projects stay on track without manual timesheet chasing. Currently in preview.
Scheduling Operations Agent (Field Service). Manages dynamic re-scheduling when disruptions occur, such as technician unavailability or travel delays, optimizing assignments in response to real-time conditions.
Quality Evaluation Agent (Customer Service). Evaluates the majority of customer interactions, human and AI-handled alike, surfacing quality insights at a scale traditional sampling-based audits cannot match.
Building Your Own Agents with Copilot Studio and MCP
Most organizations have processes specific enough that the pre-built agents above won’t cover everything. The practical path most teams follow:
- 1. Enable the Dynamics 365 ERP MCP Server through system administration settings in your F&O environment.
- 2. Open Microsoft Copilot Studio, Microsoft’s low-code platform for building agents without writing code.
- 3. Connect the MCP Server to your agent via the Tools tab, filtering for Model Context Protocol.
- 4. Define your agent’s instructions in plain language, describing the goal and boundaries.
- 5. Add knowledge sources and custom logic via SharePoint content, internal knowledge bases, AI Plugins, Virtual Entities, and Business Events for edge cases.
- 6. Test, refine, and publish using Copilot Studio’s built-in test panel before going live to a Teams bot, embedded D365 experience, or another endpoint.
Configuring the MCP connection and scoping an agent’s tool access safely is where Devsinc’s AI & Data Systems team typically works alongside a customer’s own D365 administrators, since the boundaries an agent operates within are a governance decision as much as a technical one.
Governance and Security: What You Need to Know
The most common concern teams raise is control, specifically what happens when an agent acts on data without a human in the loop at every step. Microsoft has built several layers of governance in:
- Role-based access control applies fully to agents, they cannot read tables or execute actions outside their assigned security role.
- Audit logging is built in, every agent action is recorded for compliance review.
- Human-in-the-loop controls can be configured per workflow, letting an agent surface a recommendation and wait for approval rather than act autonomously.
- Data isolation through Dataverse ensures agents only interact with explicitly granted entities.
Microsoft has also introduced Immersive Home, an AI-powered workspace in the 2026 Release Wave 1 that serves as a central landing page for managing and monitoring agents across your environment.
An agent reconciling your accounts is not working from a last-night export. It is working from the same live data your finance team sees.
What’s Coming: 2026 Release Wave 1
The 2026 Release Wave 1 (April to September 2026) includes:
- Enhanced MCP server capabilities, moving from fixed tools to a dynamic, configurable architecture
- Analytics MCP server, extending agent reasoning to structured financial and operational metrics
- AI-powered picking and stock rebalancing for warehouse operations, including hands-free scanning
- Immersive Home workspace as a central hub for agent management and oversight
- Expanded Copilot Credit model, with pooled monthly credits on Dynamics 365 Premium SKUs
Is This the Right Time to Start?
With AI agents and MCP becoming part of the platform, organizations can begin their Dynamics 365 AI adoption journey today rather than waiting for a complete transformation. The four-phase sequence practitioners recommend: Copilot enablement first, targeted agent deployment for high-volume repetitive processes second, MCP configuration for custom scenarios third, and integration of partner-built agents where gaps exist last.
Finance-specific compliance automation is a natural place to start looking for that first high-volume process. If you are already modernizing document-heavy pipelines, such as electronic invoicing in Dynamics 365 Finance & Operations, agentic AI is a logical next layer once that base automation is in place.
Key takeaways
- Agentic AI acts on goals, not fixed rules, and works with live D365 data rather than exports or snapshots.
- The Dynamics 365 ERP MCP Server is what connects agents to your ERP's data and business logic securely.
- Pre-built agents already exist for reconciliation, supplier communications, scheduling, and quality evaluation, no custom build required to start.
- Governance is built in: role-based access, audit logging, and human-in-the-loop controls let you decide how much autonomy to grant.
Learn more about our Dynamics 365 implementation approach →
References
- Microsoft, "2026 Work Trend Index: Agents, Human Agency, and the Opportunity for Every Organization."
- Microsoft, "The 2025 Annual Work Trend Index: The Frontier Firm Is Born" (April 23, 2025).
- Microsoft Learn, "Use Model Context Protocol for Finance and Operations apps."
- Microsoft Learn, "Dynamics 365 2025 Release Wave 2."