Agent Designer in Business Central: creating custom AI agents without coding

Key takeaways

• Agent Designer allows creating custom AI agents in Business Central visually — without AL coding.
• You define in natural language what the agent should do, which data it may use and when it must request human approval.
• Creation takes place in a sandbox environment. For production use the prototype can be converted into an AL extension.
• You need the AGENT-ADMIN permission for creating and AGENT-DIAGNOSTICS for monitoring and troubleshooting.
• Agent Designer is suited for standardised routine processes — for complex business logic, AL development is the better path.

How Agent Designer works

Agent Designer provides a graphical interface where you configure an AI agent step by step. You describe in natural language what the agent should do — for example: 'Monitor an email mailbox. When a complaint arrives, create a service order and assign it to the responsible technician.'

Next you define the guardrails: which tables may the agent read? Which may it write to? For which actions must it request human approval? This configuration determines how autonomously the agent operates.

Agent Designer creates a working prototype from this, which you can test in a sandbox environment. For production use you have two options: use the agent directly (for simple scenarios) or have a developer convert the prototype into an AL extension (for more robust, maintainable solutions).

Setup and permissions

Agent Designer requires Business Central Online (cloud) and is available in a sandbox environment. You need two permissions: AGENT-ADMIN (ID 9665) for creating and configuring agents, and AGENT-DIAGNOSTICS for monitoring and troubleshooting running agents.

Copilot billing must be activated in the Admin Centre, as AI processing consumes Copilot credits. The amount depends on how frequently the agent is active and how complex its tasks are.

Practical examples: what can be built with Agent Designer

Agent Designer's strength lies in the rapid implementation of standardised routine processes. Here are some realistic scenarios:

• Complaint handling: the agent monitors a service mailbox, recognises complaints and automatically creates service orders with assignment to the affected customer and item.
• Inventory alerting: the agent regularly checks stock levels for defined items and automatically creates a purchase request or notifies purchasing when levels fall below a threshold.
• Delivery status checking: the agent answers incoming email enquiries from customers about delivery status by looking up the order in Business Central and sending a status update by email.
• Dunning support: the agent identifies overdue open items and drafts payment reminders for the clerk to review before sending.

Limitations of Agent Designer

Agent Designer is a rapid prototyping tool, not a replacement for AL development. It is suited for clear, rule-based processes with manageable complexity. Where are the limits?

Complex business logic — multi-stage approval processes, conditional calculations or integrations with third-party systems — exceeds the Designer's capabilities. Converting to an AL extension is the right path here.

For high transaction volumes or time-critical processes we also recommend the AL variant: it is more performant, better testable and can be integrated into CI/CD pipelines.

Another point: agents created via the Designer are bound to the sandbox environment until converted into an extension. For production use this step is almost always advisable.

From prototype to production-ready extension

The typical path is: create a prototype in the Designer, validate in the sandbox, then have an AL developer convert it into an extension. In the extension you can refine business logic, add error handling, write tests and deploy the agent via normal extension management.

For companies without their own AL developers this is where a partner comes in. The advantage: the prototype from the Designer serves as a living specification — the developer sees exactly what the agent should do, instead of having to interpret an abstract requirements description.

Frequently asked questions

Do I need programming skills for Agent Designer?

No. Agent Designer works with natural language and a graphical interface. AL skills are only needed when you want to convert the prototype into a production-ready extension.

Can I create unlimited agents with Agent Designer?

Technically yes, but each active agent consumes Copilot credits. Costs increase with the number of agents and the frequency of their activities. We recommend starting with one or two agents and expanding gradually.

Is Agent Designer available on-premises?

No. Agent Designer and all AI agents require Business Central Online (cloud). On-premises installations cannot use this feature.

How secure are the created agents?

The agents operate within Business Central's permission model. They can only access data and perform actions allowed by their permission sets. Additionally, you define in the guardrails which actions require human approval.

Can I delete an agent created in the Designer?

Yes. In the sandbox environment you can deactivate and delete agents at any time. Agents that have already been converted into an AL extension are managed via the normal extension management.