Business Central

Business Central AI agent designer in Australia: what SMB teams should test now

A practical Microsoft Dynamics 365 Business Central guide for Australian teams reviewing the new in-product AI agent design experience, where it fits, what it still cannot do, and how to run a sensible first pilot.

Microsoft has now given that interest a new focal point. In the 2025 release wave 2 plan, Microsoft lists the in-product experience to envision and design AI agents in Business Central with public preview from 6 February 2026 and general availability in May 2026. At the same time, the feature-details section still describes the ability to design your own agents as being in public preview for some behaviour. That matters because the conversation can move beyond Microsoft’s built-in Payables Agent and Sales Order Agent into whether SMB teams should prototype their own role-specific automation, while still keeping rollout maturity under review.

The useful question is not whether every Business Central customer should suddenly build custom agents. It is whether your business now has enough product evidence to test a narrow, supervised ERP use case without confusing prototype value, production readiness, billing, and governance.

What Microsoft has officially released

• Microsoft’s release-plan page for “Envision and design AI agents in Business Central” says the feature is enabled for users by admins, makers, or analysts, was in public preview from 6 February 2026, and reached general availability in May 2026.
• Microsoft describes the experience as a low-risk environment for envisioning and prototyping custom Business Central agents using the same in-product runtime capabilities that power built-in agents.
• Microsoft says users can create and refine agents with natural-language instructions, test agent profiles and permissions, review execution logs, and experiment safely in a sandbox environment.
• The same Microsoft page also says that while the feature is listed with May 2026 general availability, the ability to design your own agents is still described as being in public preview in the feature details. That is exactly the kind of maturity nuance buyers should confirm before treating a pilot as a production commitment.
• Microsoft also says there are not yet built-in integrations in the design experience to trigger custom agents automatically from incoming emails, events, or schedules. Those scenarios can be simulated manually or implemented by professional developers through the APIs in the AI Development Toolkit for Business Central.
• On the same page, Microsoft says custom agents consume Copilot Credits as they perform their steps, using prepaid or pay-as-you-go capacity models.

Why this matters for Australian SMB teams

• This is the first Microsoft-published point where Business Central buyers can evaluate custom agent design as a real product capability, not only as partner concept work or PowerPoint ambition.
• For Australian SMB and mid-market teams, that changes the roadmap conversation from “Should we switch on Microsoft’s built-in AI?” to “Which one or two ERP workflows are mature enough to justify a controlled custom-agent pilot?”
• It also separates three decisions that are often mixed together: standard Copilot usage, built-in agent adoption, and custom-agent design. Microsoft’s own documentation makes those layers distinct, so buyers should budget and govern them separately.

What the in-product designer is good for today

• Prototyping ERP-contained scenarios where the business wants to test agent instructions, permissions, logs, and review checkpoints before committing to broader automation.
• Validating whether a repeated Business Central task can be framed clearly enough in natural language to produce consistent outputs without weakening control boundaries.
• Giving product owners, consultants, analysts, and technical teams one shared place to test instructions, assigned profiles, and execution traces against live business context in a sandbox.
• Building confidence about supervision. Microsoft says the runtime keeps actions traceable, works within assigned permissions, and supports a human-in-the-loop model for critical operations.

What it still does not solve by itself

• It is not the same as full cross-system orchestration. Microsoft explicitly separates deep in-product execution in Business Central from higher-level orchestration across products.
• It is not yet a no-code “set and forget” automation layer for mailbox, event, or scheduled triggers directly from the in-product design experience.
• It does not remove the need for process design. If the business cannot define ownership, exceptions, or approval thresholds clearly, a custom agent will surface that ambiguity rather than fix it.
• It does not make cost irrelevant. Microsoft says custom agents consume Copilot Credits, so pilot economics still need explicit review.

When to stay inside Business Central and when to step into Copilot Studio

• Stay in the in-product design experience first when the use case is tightly inside Business Central, the process owner wants traceability in the ERP context, and the team needs a low-risk way to test instructions and permissions in sandbox.
• Move into Copilot Studio when the use case needs broader channels or orchestration across products. Microsoft’s current Copilot Studio documentation says agents can connect to Business Central through the Business Central Connector or the Business Central MCP Server and then be published to channels such as live websites, Microsoft 365 Copilot, Teams, and other messaging surfaces.
• Microsoft says the connector is better for standard low-code integration and automation, while the MCP server is better for AI-agent scenarios that coordinate multiple steps. That is a useful architecture filter for buyers trying to avoid over-engineering the first pilot.
• Microsoft also says Copilot Studio requires its own user licence with available Copilot Credits capacity. So the moment a Business Central pilot expands into cross-system agent design, the commercial model changes again.

What to test before approving a pilot

• Use-case boundary. Choose one process that is repetitive, bounded, and already reasonably disciplined. Examples include summarising stock exceptions for a planner, preparing first-pass follow-up on order-hold cases, or drafting a controlled data-quality review for item or customer records.
• Permission design. Test which Business Central profiles and permissions the agent really needs. Microsoft’s documentation emphasises permission-bound execution, so least-privilege design should be part of the pilot, not an afterthought.
• Review and auditability. Use the agent logs and timeline views to check whether business users can understand what the agent did, why it did it, and where manual review must stay mandatory.
• Data quality. Poor customer, item, vendor, or document data will weaken output quality quickly. Do not judge the agent only on a clean demo dataset.
• Consumption economics. Estimate likely Copilot Credit use for the pilot and decide what volume or usage threshold triggers review before anyone calls the test a success.
• Environment governance. Microsoft’s Copilot and agent controls page says administrators can deactivate capabilities, grant or deny user access, and use data-governance controls. It also notes that some environments may require allowing data movement across geographies if the Business Central environment and Azure OpenAI geography differ.
• Online-only fit. Microsoft says Copilot and agents are available only to Business Central online customers, so hybrid or on-prem assumptions should be challenged before pilot planning starts.

Where teams usually get this wrong

• They start with a glamorous use case instead of a governed one. The first pilot should prove control and clarity, not marketing value.
• They assume “Copilot included” means custom-agent rollout is free. Microsoft’s agent documentation does not support that assumption because custom agents consume Copilot Credits.
• They treat agent design as a substitute for process ownership. If no one owns the data, approvals, or exception path, the pilot will generate debate instead of evidence.
• They jump too quickly from in-product prototyping to cross-system orchestration without noticing the architecture, licensing, and support model have changed.

A sensible first pilot for an Australian SMB team

• Start in one sandbox company with one process owner, one analyst or admin, and one narrow scenario that lives mostly inside Business Central.
• Run the agent manually through a small batch of realistic cases, including messy data and at least a few exceptions, because Microsoft says automatic triggers are not yet built into the in-product designer itself.
• Measure three things only: output usefulness, review effort, and Copilot Credit consumption. If the team cannot explain those clearly after the pilot, it should not broaden scope yet.
• Decide the next path explicitly: keep refining in-product instructions, hand the scenario to developers using the AI Development Toolkit, or redesign it as a broader Copilot Studio workflow if cross-system orchestration is genuinely required.

What Australian buyers and current customers should conclude now

• Microsoft has made custom-agent design concrete enough to justify a structured pilot in Business Central, especially for online SMB teams already evaluating AI but not yet ready for a big automation programme.
• The strongest early use cases are controlled, ERP-contained tasks where permissions, logs, and human review matter as much as speed.
• The wrong conclusion is that May 2026 availability makes Business Central an instant autonomous-workflow platform. The better conclusion is that Microsoft now offers a safer path to test custom agents, provided the business keeps scope narrow and governance explicit.

FAQ

• Is the Business Central AI agent designer relevant to Australian teams right now? Yes, as a roadmap and pilot topic. Microsoft’s release-plan page lists May 2026 general availability, but the same feature details still use some preview language, so Australian teams should treat it as current enough to test and verify rather than current enough to assume broad production maturity everywhere.
• Can non-developers use it? Partly. Microsoft says the experience is enabled for admins, makers, or analysts and supports natural-language instruction design, but production-style triggering or deeper integration can still require developer work.
• Does this work for Business Central on-premises? No. Microsoft’s current Copilot and agents overview says Copilot and agents are available only to Business Central online customers.
• Are custom agents included in the normal Business Central licence? Not in the same way as standard Copilot. Microsoft says built-in Copilot comes with the Business Central licence, while custom agents consume Copilot Credits.
• Should the first pilot be built in Copilot Studio instead? Usually not. Start in-product if the scenario is mainly inside Business Central. Move to Copilot Studio only when the use case genuinely needs external channels or cross-system orchestration.

Sources used

• Microsoft Learn release-plan page: Envision and design AI agents in Business Central.
• Microsoft Learn release-plan overview: Copilot and agents in Business Central.
• Microsoft Learn documentation: Configure Copilot and agent capabilities in Business Central.
• Microsoft Learn documentation: Create agents in Copilot Studio that connect to Business Central.