How to Build AI Agents in Dynamics 365 Business Central: A Partner's Practical Guide (2026)

With Business Central 2025 Wave 2, Microsoft has made custom AI agent development available to all partners and developers through the Business Central AI Development Toolkit. This blog is for Dynamics 365 Business Central partners, AL developers, and functional consultants who want a practical, step-by-step understanding of how to design, test, and deploy custom agents.

At Buy Business Central, powered by Cetas, a Microsoft Solutions Partner with 16+ years of Business Central expertise, we have been working closely with these tools since preview. Here is what we have learned.

What Is the Business Central AI Development Toolkit?

The Business Central AI Development Toolkit is Microsoft's in-product environment for building custom AI agents inside Dynamics 365 Business Central. Available as a public preview, it allows you to prototype agents in natural language, test them in a sandbox, and inspect execution traces before any production deployment.

This is distinct from the built-in Sales Order Agent and Payables Agent. Those are Microsoft-developed, ready-to-use automations. The AI Development Toolkit lets you build agents tailored to your specific business workflows.   

How to Build a Business Central AI Agent: A Three-Step Process

Step 1: Write the Agent's Instructions - This Is Where Most Projects Succeed or Fail

Every Dynamics 365 Business Central AI agent begins with a mission statement, written in plain language. The agent does not follow a script. It interprets a goal and determines its own steps to achieve it. That makes instruction writing a discipline in itself. Effective agent instructions define:

• The agent's specific purpose (tight scope = reliable behaviour)
• The conditions under which it must pause for human approval
• The data it is authorised to read and write
• What it must never do, regardless of input

Microsoft's toolkit includes a library of instruction keywords, structured signals like PROHIBITED, MUST, REQUIRED, and IF-THEN-ELSE, that give the agent's underlying model precise guardrails while keeping instructions in natural language. Getting this layer right has a larger impact on agent reliability than any technical configuration downstream.

Practical tip: Write a separate "edge case" document alongside your instructions. For every main instruction, ask: "What would a misinterpretation of this look like?" Test those exact misinterpretations in the sandbox before calling the agent complete.

Step 2: Use the Iteration Environment - BC's Agent Debugging Loop

The sandbox iteration environment in the Business Central AI Development Toolkit is one of the most valuable tools for agent development. It logs the agent's complete execution trace: every page it navigated, every field it read, every decision branch it evaluated, and every place it asked for human input.

This is not just for debugging. It is the mechanism for responsible deployment. Before any agent is used in a live Business Central environment, it should be stress-tested against:

• Incomplete or missing master data
• Multi-currency and multi-company scenarios
• Orders that partially match business rules (the hardest edge case)
• Deliberate attempts to trigger out-of-scope behaviour

Functional consultants play a critical role here, not writing code, but designing test scenarios and evaluating whether the agent's reasoning matches actual business intent.

Step 3: Go Further with the AL Agent SDK

Once an agent design is validated through the no-code interface, AL developers can codify and extend it using the AL Agent SDK, Microsoft's programmatic layer for Business Central AI agent development. The SDK enables:

• Programmatic agent registration: Define the agent's identity, instructions, and permission scope entirely in AL, making it a deployable extension
• Task API access: Trigger agents from external events like incoming emails, webhooks, and scheduled jobs, not just manual user actions
• Setup page integration: Build a dedicated configuration UI for admins, using the ConfigurationDialog page type
• Version control: Because the agent is an AL extension, it lives in your source repository alongside your other Business Central code

This is where the Business Central AI agent becomes a production-grade, supportable, upgradeable product and not just a sandbox prototype.

Learn from Microsoft's Reference Implementation

The Sales Validation Agent: A Blueprint for Production-Ready Agent Design

Microsoft's Sales Validation Agent sample (available in the BCTech GitHub repository) is the single best learning resource for Business Central AI agent development right now, and it is significantly underutilised by the partner community.

It demonstrates a complete, well-scoped agent that reviews incoming sales orders for completeness and policy compliance. More importantly, it shows the design patterns that make agents reliable in production:

• Tight scope: The agent does exactly one thing and does it well
• Minimal permissions: The agent's permission set gives it access only to what it needs
• Structured human handoff: Approval workflows are wired in at the right decision points, not everywhere and not nowhere

These three patterns, tight scope, minimal permissions, and structured handoff, are the reusable architecture of good Business Central AI agent design. They should inform every custom agent you build.

Beyond Business Central: MCP, Copilot Studio, and Microsoft 365

Business Central is implementing support for the Model Context Protocol (MCP), Microsoft's open standard for enabling AI assistants to securely consume tools and data from external systems.

In practice, this means a custom AI agent you build in Business Central today can be surfaced as an MCP tool, callable by Declarative Agents in Microsoft Copilot Studio, embedded in Teams, or orchestrated across Microsoft 365 workflows.

The Sales Order Agent you build in BC becomes one component of a larger intelligent workflow: an email arrives in Outlook, a Copilot agent in Teams interprets it, and a Business Central agent creates and validates the order, all connected through MCP without custom middleware.

This composability is not speculative. It is the architectural direction Microsoft is actively building, and the Business Central AI Development Toolkit is the entry point for partners.

Frequently Asked Questions

1. Do I need to write AL code to build an AI agent in Business Central?

No. The Business Central AI Development Toolkit allows you to prototype and test agents using plain-language instructions, no AL required. For production deployment, developers can optionally use the AL Agent SDK to codify the agent as a formal extension.

2. How is the Business Central AI Development Toolkit different from Copilot Studio?

Copilot Studio is Microsoft's low-code platform for building conversational agents across the Microsoft ecosystem. The Business Central AI Development Toolkit is purpose-built for Business Central, where agents reason directly over BC data and interact with BC pages via the logical UI API. The two can work together, as BC agents can be surfaced as MCP tools callable from Copilot Studio.

3. Is the AL Agent SDK the same as the AI Development Toolkit?

They are complementary. The AI Development Toolkit includes the no-code Agent Design Experience for prototyping. The AL Agent SDK is the developer layer that lets you register, configure, and extend agents programmatically in AL code.

Where to Start with Business Central AI Agent Development

Building AI agents in Business Central is no longer experimental. The toolkit is production-ready, the design patterns are proven, and the partners who invest in this knowledge now will be the ones clients call first when agentic AI moves from preview to standard practice. Here is where to start today:

• Open a BC sandbox and activate the Agent Design Experience
• Study the instruction keyword library in Microsoft Learn before writing a single instruction
• Clone the Sales Validation Agent from GitHub and run it before modifying anything
• Design a full edge-case test matrix before declaring any agent production-ready
• Think about MCP composability from day one and scope agents with interoperability in mind

New to Business Central AI agents? Start with our foundational guide here.

If you are ready to go further and explore what custom BC agents could look like for your specific workflows, our team of certified Dynamics 365 Business Central consultants can help you move from proof-of-concept to production.

With 16+ years of Business Central implementation experience and a deep technical bench covering AL development, Copilot integration, and Power Platform, we are the right Microsoft Dynamics 365 Business Central Partner to speak to. Talk to our team at sales@buybusinesscentral.com