Custom AI agents in Dynamics 365 Business Central: Benefits, limitations, and best use cases 

Artificial intelligence is becoming an increasingly important part of Microsoft Dynamics 365 Business Central, giving organizations new ways to automate work, analyze information, and support decision-making. One emerging option is the ability to design and build custom AI agents within Business Central.

In this article, “Business Central agents” refers specifically to custom AI agents created within Business Central to support defined ERP workflows. These agents are different from Microsoft-delivered agents, such as the Payables Agent, and different from agents created in Copilot Studio that may connect to or be triggered by Business Central.

Custom Business Central agents can help organizations create AI-assisted workflows that review Business Central data, prepare recommendations, summarize information, and support users through structured business processes. They are especially useful when an organization wants AI to assist with a defined process while keeping human review, security, and ERP governance in place.

However, custom agents are not a one-size-fits-all automation tool. They have important limitations around deployment, security, prompt design, consistency, testing, and long-term support.

After working extensively with custom Business Central agents, Rand Group has identified clear patterns around where they provide practical value and where more traditional automation or hybrid AI solutions may deliver better long-term results. This article explores how custom agents work, where they fit, where they fall short, and how to decide whether they belong in your Business Central automation strategy.

What are custom Business Central agents?

Custom Business Central agents are AI-assisted workflows that can be created within Microsoft Dynamics 365 Business Central to help users complete defined business processes. They are designed to work with Business Central data, follow structured instructions, and support tasks such as reviewing records, summarizing information, identifying exceptions, and preparing recommendations for user review.

Microsoft introduced this custom agent functionality for Business Central as part of the 2026 release wave 1, expanding how organizations can apply AI directly within ERP workflows. Instead of relying only on standard automation or broad AI tools, businesses can create agents that are more closely aligned with their Business Central processes, data, and user roles.

In practical terms, a custom Business Central agent might review a group of records, identify items that need attention, summarize potential issues, and prepare information for a user to approve. The agent supports the process, but it does not replace the ERP system, business rules, security permissions, or human oversight.

For example, a custom agent could help review overdue customer balances, summarize invoice exceptions, or prepare records for approval. In each case, the agent helps users work faster by organizing information and suggesting next steps, while the final decision remains with a person.

This article focuses on custom Business Central agents created within Business Central.

Where custom Business Central agents deliver the most value

Custom Business Central agents perform best when organizations need AI to assist with decision-making rather than fully automate a process from beginning to end.

Their strongest use cases are structured workflows where an agent reviews Business Central data, prepares a recommendation, and presents that recommendation to a user for review before any action is taken. This approach allows organizations to benefit from AI while maintaining appropriate human oversight.

Examples include:

• Reviewing transaction data before posting
• Identifying records that may require attention
• Preparing recommendations for approval
• Summarizing information for user review
• Supporting multi-step workflows that include human validation
• Helping users evaluate exceptions, anomalies, or incomplete records

These human-in-the-loop scenarios are especially important in ERP environments, where decisions often affect financial results, inventory, customer records, compliance, or operational performance. In these situations, AI can help accelerate analysis, but the final decision should still remain with the appropriate user or approver.

Business Central agents are also well suited for lower-effort workflows where organizations want to introduce AI capabilities without starting with a large development project. Because the tools are built into the Business Central experience, teams can begin experimenting with agents, test ideas in a sandbox, and evaluate whether a workflow has enough value to justify further investment.

This makes custom Business Central agents a useful starting point for organizations beginning their AI journey within ERP.

Why custom Business Central agents are useful for AI experimentation

One advantage of custom Business Central agents is that they give organizations a way to experiment with AI-assisted workflows directly in a Business Central context.

Teams can begin by testing a defined process, evaluating how an agent interprets Business Central data, and determining whether the use case is valuable enough to justify further investment. This can be a practical step for organizations that want to explore AI without immediately committing to a larger Azure AI or custom development project.

However, experimentation is different from production readiness. Moving a custom agent into a production environment requires more planning around AL development, permissions, testing, deployment, support, and governance.

Understanding the current limitations

Custom Business Central agents are capable tools, but they intentionally simplify many aspects of AI implementation. That simplicity is helpful for adoption, but it also means organizations have less control than they would with a fully custom AI solution.

Common limitations include:

• Limited control over which AI model is used for specific tasks
• Limited visibility into how responses are generated
• Permission and security considerations that require careful configuration
• Potential variability between identical executions
• Less flexibility than custom AI implementations
• A need for careful testing before production use

The inability to choose different AI models is especially important for advanced scenarios. Different large language models may perform better at different types of tasks, such as summarization, classification, structured data extraction, reasoning, or natural language interaction. Business Central agents provide a more simplified experience, but that can limit control for organizations with complex requirements.

Organizations also need to pay close attention to security. Each Business Central agent should be treated with the same seriousness as any other user or process that can access business data. Permissions should be carefully defined so the agent can only access the records and actions needed for its intended purpose.

Improper configuration could allow an agent to access information beyond its scope or interact with data in ways the organization did not intend. This is why security reviews should be part of any production deployment, especially for workflows involving financial records, customer data, vendor information, approvals, or posting routines.

Organizations should also expect some variability in AI responses. Even when processing similar data with similar instructions, AI-generated responses may not always be identical. This is common with generative AI, but it reinforces the importance of using agents in workflows where human review, validation, and exception handling are built into the process.

Understanding these limitations helps organizations deploy Business Central agents where they can provide value without introducing unnecessary operational risk.

Why prompt quality matters

As with many AI solutions, the quality of the result depends heavily on the quality of the instructions provided.

Custom Business Central agents require clear, detailed, and well-structured prompts to produce useful outputs. Vague instructions often lead to incomplete recommendations, inconsistent results, or responses that require additional user interpretation.

In practical testing and implementation scenarios, organizations may encounter situations where agents:

• Truncate long lists of information
• Miss records because filters or data inputs are incomplete
• Produce inconsistent summaries across multiple runs
• Require additional data preparation before processing
• Need more explicit instructions to produce usable recommendations

These issues do not mean Business Central agents are ineffective. Rather, they highlight the importance of prompt design, data preparation, and testing. Organizations should not assume that an agent will understand business context without being given enough structure.

For example, an agent asked to “review overdue invoices” may produce a general summary. A more useful prompt would specify which fields to review, how to define overdue status, whether to group results by customer or aging bucket, what exceptions to flag, and what type of recommendation the user expects.

In some cases, AL development can improve performance by preparing the data before it is passed to the agent. Developers can consolidate records, apply filters, structure data, or ensure the complete dataset is available for review. This can help the agent produce more consistent and useful outputs.

Without this supporting structure, organizations may find that agent performance varies depending on the volume, complexity, and quality of the data being analyzed.

The hidden dependency many organizations overlook

Custom Business Central agents are not simply a no-code feature that business users can move into production on their own. While teams may be able to design, test, and refine agent ideas in a sandbox, production-ready custom agents typically require AL development support.

AL is the development language used to customize and extend Business Central. For custom agents, AL may be needed to define the agent, package the solution, prepare data, manage permissions, support deployment, and ensure the workflow behaves consistently in a production environment.

This creates a dependency organizations should plan for early. Instead of treating custom agents as entirely self-service, teams should approach them like other Business Central enhancements: with clear ownership, technical review, testing, documentation, deployment planning, and long-term support.

Production planning should include:

• AL development support
• Production deployment assistance
• Security and permissions review
• User acceptance testing
• Governance and approval processes
• Ongoing maintenance and updates
• Documentation for future support

This requirement should not discourage organizations from exploring Business Central agents. However, it does mean AI initiatives should be planned like any other ERP enhancement. Teams should consider ownership, supportability, testing, deployment timing, and change management before moving an agent into production.

When traditional automation is the better solution

One of the most common implementation mistakes is using AI where standard automation would produce better results.

Not every repetitive process requires artificial intelligence. If a workflow is deterministic, follows consistent business rules, and produces predictable outputs, traditional Business Central customization is often the better choice.

Examples include:

• Data validation
• Scheduled processing
• Rule-based approvals
• Record updates
• Posting controls
• Standard business logic
• Repeatable calculations
• Fixed exception rules

These scenarios are typically better handled through AL code or standard Business Central configuration rather than AI agents.

Traditional automation is predictable, repeatable, and easier to test against known outcomes. It also does not consume AI processing each time it executes. While custom development requires an initial investment, the resulting automation may be more reliable and cost-effective for processes that do not require reasoning or interpretation.

For example, if an organization needs to block posting when a required field is missing, AI is unnecessary. A validation rule or customization can enforce that requirement consistently every time. If an organization needs to summarize why a customer account appears risky based on several data points, an AI-assisted workflow may be more appropriate.

The key is to distinguish between workflows that require reasoning and workflows that simply require automation.

A hybrid approach often delivers the strongest results

For more sophisticated business scenarios, a hybrid approach may provide the best balance of control, reliability, and intelligence.

Rather than relying entirely on Business Central agents, organizations can use AL development to manage structured business logic while invoking AI only for the parts of the workflow that benefit from interpretation, summarization, classification, or reasoning.

The right approach depends on whether the workflow requires AI interpretation, fixed business rules, or a combination of both.

In this model, AL code manages the parts of the workflow that need to be predictable, secure, and repeatable. This may include preparing data, applying filters and validation rules, enforcing security controls, managing business logic, orchestrating workflow steps, and preserving auditability and transactional consistency.

AI is then used only for the parts of the process that benefit from interpretation or reasoning. For example, it can summarize complex information, classify records, draft recommendations, interpret unstructured text, identify potential exceptions, or help users review large volumes of data more efficiently.

This hybrid architecture gives organizations greater control over the overall workflow while still allowing them to benefit from AI where it adds value. Since deterministic automation handles the structured parts of the process, results are often more consistent and predictable. Organizations also gain better visibility into what information is sent to AI, more flexibility when selecting AI capabilities, and stronger governance for production environments.

Although this approach requires development expertise, organizations considering production AI implementations often need that expertise regardless of the solution they choose. The goal is not to use AI everywhere, but to apply it where it produces a measurable improvement in speed, insight, accuracy, or user productivity.

Governance considerations for custom Business Central agents

Governance is an important part of any AI initiative, especially in an ERP system where data and processes are business critical.

Before deploying Business Central agents into production, organizations should define clear policies around how agents are created, tested, approved, and monitored. Without governance, teams may create AI workflows that are difficult to support, inconsistent across departments, or misaligned with security requirements.

Key governance questions include:

• Who is allowed to create or modify agents?
• What data can each agent access?
• What actions can the agent recommend or perform?
• Who reviews and approves agent outputs?
• How are agent results tested before production deployment?
• How are changes documented?
• How will the organization monitor performance over time?
• What happens if an agent produces an incomplete or incorrect recommendation?

These questions are especially important for workflows involving finance, compliance, approvals, customer data, or operational decisions. Organizations should also consider whether agent outputs need to be logged, reviewed, or included in audit processes.

Governance does not need to slow down innovation. In fact, clear governance often makes AI adoption easier because users understand where agents can be used safely and where additional review is required.

Choosing the right solution for your business

Custom Business Central agents are an exciting addition to Microsoft’s AI ecosystem, but success depends on applying them to the right business problems.

Organizations should view custom agents as one option within a broader automation strategy rather than the default answer for every workflow.

In general:

• Choose custom Business Central agents when human review is an important part of the process.
• Use traditional AL development for repetitive, rule-based automation.
• Consider hybrid AI solutions when workflows require both deterministic business logic and advanced AI reasoning.
• Apply governance and security reviews before moving agents into production.
• Test thoroughly with realistic data before relying on agents for business-critical processes.

A practical evaluation should begin with the business process, not the technology. Organizations should first define the problem they are trying to solve, the decision being made, the data required, the acceptable level of risk, and the expected outcome. For organizations still developing their broader AI strategy, these guidelines for Dynamics 365 can help frame the planning process. From there, they can determine whether a Business Central agent, standard automation, custom development, or a hybrid AI solution is the best fit.

Taking this strategic approach helps organizations maximize return on investment while avoiding unnecessary complexity.

Why partner with Rand Group

Successfully adopting AI within Microsoft Dynamics 365 Business Central requires more than enabling new features. It requires the right balance of process understanding, technical architecture, and practical implementation experience.

Rand Group helps organizations move beyond experimentation with Business Central agents and into real, production-ready solutions that align with business goals. Our team brings together both functional consultants and AL developers who understand how Business Central works at the core, as well as how AI capabilities should be applied in a controlled and sustainable way.

We work across the broader Microsoft ecosystem, including platforms such as Microsoft Dynamics 365 Business Central and Microsoft Dynamics 365 Finance and Operations, helping organizations determine where each solution fits within their enterprise architecture. This includes designing integrations with cloud services such as Microsoft Azure and applied AI capabilities from Microsoft AI and Machine Learning.

Our approach also considers the people side of technology adoption. Successful implementations depend on effective training, change management, and adoption strategies, which is why we emphasize structured enablement aligned with Microsoft User Adoption principles.

Ultimately, our goal is to help organizations make confident decisions about when to use Business Central agents, when to rely on traditional AL development, and when a hybrid AI approach delivers the most value.

Frequently asked questions about Dynamics 365 Business Central custom agents

Next steps 

If your organization is evaluating custom AI agents within Microsoft Dynamics 365 Business Central, the next step is to assess where this specific approach can create meaningful value within your existing ERP processes.

For some workflows, custom Business Central agents may be a strong fit—particularly where structured review and human approval are required. In other cases, traditional AL-based automation or a hybrid approach using Azure AI may provide greater consistency, control, and cost efficiency.

Rand Group can help you evaluate your current Business Central environment, identify suitable AI and automation opportunities, and determine the right implementation approach based on your business requirements and technical landscape. This includes guidance on custom Business Central agents, AL development, Microsoft-delivered agents, Copilot Studio, and Azure AI integration strategies.

If you are considering how AI fits into your Business Central roadmap, contact Rand Group to discuss your goals and help you build a clear, practical path forward.