← Back to TutorialsAI + BUSINESS CENTRAL

AI in Business Central: Complete Beginner Guide

Dynexal • Beginner to Intermediate • 15 min read

Artificial intelligence is becoming a practical part of Microsoft Dynamics 365 Business Central. This guide explains Copilot, generative AI, AI-powered business scenarios, AL development with AI, Azure OpenAI, AI agents and the Business Central MCP approach.

What you will learn: AI basics, Copilot, practical use cases, AI extension architecture, Azure OpenAI, AL examples, AI agents, MCP, security, responsible AI and project ideas.

What does AI mean in Business Central?

AI in Business Central is not just a chatbot. It can assist users with tasks, generate or summarize content, analyze information and support business processes. Microsoft describes Copilot as an AI-powered assistant that works collaboratively with users, while agents are designed to handle specific tasks with more autonomy and keep work transparent and reviewable.

The important idea for developers is that AI can become part of an ERP workflow instead of living as a separate application.

Copilot vs AI agents

Copilot: assists a user while the user initiates or reviews the work.
AI agent: can perform defined tasks with minimal human input, subject to the permissions and configuration provided by the solution.

User
  │
  ├── Copilot → Suggest / Summarize / Assist
  │
  └── Agent   → Plan / Execute defined tasks
                    │
                    ▼
             Business Central
                    │
          Customers / Items / Orders
                    │
                    ▼
               Business data

For production systems, the distinction matters because an agent may need permissions to create, modify or execute business operations, while a Copilot experience can be designed around suggestions and human review.

Practical AI use cases

1. Finance

AI can help users understand financial information, summarize records and assist with repetitive finance-related tasks. Always validate generated results before posting or making financial decisions.

2. Sales and customer service

AI can help draft customer-facing content, summarize customer information and support sales workflows.

3. Inventory and supply chain

AI can assist users in understanding inventory information and identifying patterns or exceptions that deserve attention.

4. Company setup

AI-assisted experiences can reduce repetitive setup work and make Business Central easier for new users to learn.

5. Developer productivity

AI coding tools can help developers understand AL symbols, build projects, diagnose errors, debug extensions and work through development tasks faster. Human review and normal AL testing remain important.

How AI fits into an AL extension

Microsoft provides developer tools for Copilot that allow AL developers to integrate generative AI capabilities into Business Central extensions. The System.AI namespace includes functionality for connecting extensions to Azure OpenAI, and Business Central also provides Copilot-oriented UI patterns such as the PromptDialog page type.

A typical custom AI extension can be structured like this:

Business Central Page
       │
       ▼
  User Input
       │
       ▼
  AL Codeunit
       │
       ├── Validate input
       ├── Build prompt
       ├── Add relevant BC context
       │
       ▼
   AI service
       │
       ▼
  Validate response
       │
       ▼
  Show suggestion / result
       │
       ▼
 Human review or approved action

Azure OpenAI and Business Central

Business Central supports AI development through Azure OpenAI. Microsoft's current developer toolkit is designed to help partners build AI-powered extensions and Copilot experiences. The toolkit includes an AI module and APIs for working with Azure OpenAI from AL.

When developing your own AI capability, you need to consider the Azure resource, model deployment, authorization, region availability, consumption and data governance requirements. Microsoft's setup guidance covers the Azure subscription, access, resource creation and model deployment steps.

Important: never put API keys, passwords or other secrets directly into AL source code or a public Git repository. Use an appropriate secure configuration and secret-management approach.

Simple AL concept

The following example is intentionally simplified to demonstrate the shape of an AI-enabled AL solution. Production implementations should use the current System.AI APIs, authorization model and error handling documented for your Business Central version.

codeunit 50100 "AI Product Assistant"
{
    procedure GenerateSuggestion(ProductText: Text): Text
    begin
        // 1. Validate the input
        // 2. Configure the supported AI service
        // 3. Send a controlled prompt
        // 4. Validate the response
        // 5. Return a suggestion for user review
    end;
}

The example is deliberately not tied to a specific model API signature because Microsoft evolves the AI toolkit. For real development, use the System.AI APIs available in the target Business Central version.

Prompt design for Business Central

A good enterprise prompt should be specific about the task, context, output format and limitations.

Role: Business Central assistant
Task: Summarize the sales order exception
Context: {{approved business data}}
Rules:
- Do not invent missing values.
- Return only information supported by the context.
- Clearly identify uncertainty.
Output:
1. Issue
2. Evidence
3. Suggested next step

Developer tip: do not send unnecessary business data to an AI service. Select only the context required for the task and enforce Business Central permissions before retrieving that data.

Grounding AI with Business Central data

AI responses become more useful when they are grounded in current business information. Business Central's AI experiences can retrieve relevant data using normal Business Central access mechanisms, so the AI experience should not be designed as a way to bypass permissions.

For custom extensions, think carefully about which records, fields and calculated values are included in the AI context. The safest design is to expose only the minimum data needed to complete the task.

Business Central MCP and AI agents

Model Context Protocol, or MCP, provides a standardized way for AI applications to discover and use tools and data. Business Central's MCP server can expose Business Central data and operations to supported AI clients.

By default, exposed API pages can provide read access. Write operations such as creating, modifying or deleting data require explicit configuration and permissions. This makes API design and permission management important when building agent workflows.

AI Client / Agent
       │
       │ MCP
       ▼
Business Central MCP Server
       │
       ├── API Pages
       ├── Business data
       └── Allowed actions
              │
              ▼
        Business Central

Security first: giving an AI agent write access is fundamentally different from giving it read access. Define the smallest possible set of operations, permissions and APIs required for the business scenario.

AI agents for AL development

Modern Business Central development also includes AI-agent tooling for AL development. Microsoft's current tooling can allow AI agents to build projects, publish extensions, download symbols, search symbols, retrieve diagnostics and work with debugging workflows depending on the development surface.

This does not remove the need for an AL developer. Developers still need to review generated code, understand business requirements, validate security, run tests and decide whether a proposed change is appropriate for the solution.

Responsible AI checklist

Keep a human review step for business-critical generated results.
Use least-privilege permissions for AI and agent operations.
Send only the data required for the task.
Do not expose secrets in prompts, code or logs.
Validate AI output before posting financial or operational documents.
Log important AI-driven actions so they can be investigated.
Test failure cases, incomplete data and unexpected model output.
Check regional, licensing and feature-availability requirements.

AI project ideas for Business Central developers

AI Item Description Assistant: generate draft item descriptions for user review.
Sales Order Summarizer: summarize order risks, missing information and exceptions.
Customer Query Assistant: retrieve approved customer information and explain it conversationally.
Invoice Email Assistant: draft customer communication from approved invoice data.
Inventory Insight Assistant: explain stock exceptions using controlled inventory context.
AI Developer Assistant: combine AL symbol search, diagnostics and coding assistance into a developer workflow.
Business Central MCP Agent: build a controlled agent that reads selected ERP data and performs explicitly permitted actions.

Standard Copilot, custom AI extension or MCP?

Requirement	Good starting point
Use an existing Microsoft AI experience	Standard Business Central Copilot
Add a specialized AI workflow to an extension	AL + Developer Tools for Copilot
Let an AI client interact with exposed BC tools/data	Business Central MCP Server
Need a completely different AI architecture	Custom API / Azure integration

Recommended learning path

Learn AL fundamentals, tables, pages and codeunits.
Understand Business Central APIs and integrations.
Learn basic prompt engineering and generative AI concepts.
Explore the Business Central AI and Copilot capabilities.
Learn the System.AI developer toolkit and PromptDialog patterns.
Build a small AI extension with human review.
Study permissions, data governance and responsible AI.
Explore AI agents and MCP after understanding APIs and security.

Continue learning:

Business Central API Integration →
Codeunits in Business Central →
Page Extensions in Business Central →
Shopify and Business Central Integration →
AL Tables in Business Central →

This guide is an educational overview. AI features, developer APIs, availability and configuration can change between Business Central versions. Always verify the current Microsoft documentation for your target release before implementing production solutions.