Designing Agents for Business Central

ajkauffmannBusiness Central9 hours ago28 Views

Last year, at BCTechdays 2025, Microsoft announced the Business Central Agent Builder Playground. Back then, it was in private preview. So only selected partners could play with it. Later that year, in November at Directions EMEA 2025 in Poznan, Poland, the Agent Builder Playground was again demonstrated. And in February 2026, it made it to v27.4 as a preview version. But now renamed to Envision and design AI agents in Business Central. Or shorter: Agent Design Experience.

Links to all information, including the YouTube playlist, are at the end of this post.

In this post, I’m not going to dive into the details of how this agent design experience works. Instead, I want to discuss what it actually is and what it is not. Because people, including myself, started to ask questions about this experience and what exactly the possibilities are. I will also briefly cover how to convert a configured agent into a production-ready agent.

Here is a screenshot from the introduction video on YouTube.

When you combine the audience (consultants, product owners, specialists, makers) with the capability ‘Create new agents’, people expected this to be a tool for creating and running low-code agents in Business Central. So we went in, created an agent with instructions, played with it until it worked, and then started looking at how to deploy it. How can a task be triggered from a business process, like releasing or posting a document? And we asked questions like, “Why do we need to make 5 clicks to create a task?” Why can we only see running and finished tasks in the task pane, and not create new tasks from there? And how can we get this into production?

And honestly, this confusion is not entirely due to Microsoft’s messaging not being clear. It was more of us, not closely paying attention to the message, but immediately jumping to conclusions. Ok, the bullet ‘Create new agents’ caught attention and set the expectation. And we didn’t really understand what was meant by ‘design proof of concept agents’.

So, what is it not?

A low-code agent
Triggered by any internal or external event
Deployable to a production environment

Let’s look at what it actually is

This new agent designer should be used as a toolbox to simulate an agent. You can:

Write, test, and refine the instructions
Simulate triggering by manually entering the instructions, including attachments
Monitor and review the steps
Experience the human-in-the-loop from the task pane
Test with different profiles and permissions
Troubleshoot the executed code by attaching a debugger (obviously for developers)
Review consumption of AI credits

And all of that in a sandbox environment, so as not to risk affecting production data.

You get the picture? It’s more or less similar to the in-client page designer. With one difference, though. The in-client page designer actually creates an app that can be downloaded as a starting point or integrated into an existing app. That’s not what the agent designer provides. At least not yet.

Converting to AL code

So, what should you do when you have designed an agent and trust it to work on production data? You have to convert the agent to AL code. That’s not just a click on a button. You need to take a few steps.

The steps for converting an agent to AL code are:

Export the agent. This is an XML file containing all the design details.
Create a new project in VS Code with AL: New Project and choose the template Agent.
Follow the instructions in the README file to apply the configuration
Write code to trigger the agent by sending task messages

Wait… is there no way to just import that XML file and get a read-to-go Agent? No, unfortunately not. However, we have coding agents. A simple prompt can help us to apply the exported XML file and customize the AL code.

The same prompt in plain text:

This workspace contains the boilerplate code for an AL agent.
Instructions to customize the code are in #file:README.md
Further instructions how to write AL agents can be found at Microsoft Learn: https://learn.microsoft.com/ca-es/dynamics365/business-central/dev-itpro/ai/ai-development-toolkit-landing-page. This is a landing page with navigation links to detailed information.
The agent configuration is available in #file:SALES VALIDATION-agent.xml
Customize the code according to the instructions and the configuration file.

This is a screenshot of what happens next. Be prepared that this process can take up to 15 minutes to complete (or longer, depending on your configuration).

The final result is an agent app that compiles and can be installed and tested right away.

Tips

After publishing the agent to a sandbox, you will find that the actions Edit instructions and Export agent definition on the Agent Card are disabled. That is because these actions are added by the app Agent Designer Experience. This app includes pages to edit the instructions and extends the Agent Card. But the actions are only available for agents created by the agent designer. If you want to allow users to view or edit instructions, you need to include an edit instructions page in the agent app. I hope Microsoft moves the edit instructions page from the designer app to the agent framework, so we can benefit from it and have a uniform user experience.

Another thing to look at is the agent setup page. This is a special page of type ConfigurationDialog. In the template project, the setup page has only one group. If the agent needs extra settings, then you should add more groups to the setup page.