Skip to main content
What if you could turn a short prompt into a working JTL Cloud App, complete with access to real ERP data, without manually writing the application code?That’s what Tobias Lengsholz, a software engineer at JTL, set out to demonstrate.The task was straightforward: build a customer-value dashboard that shows purchase history directly from a customer’s record in On-Prem ERP. Starting with a prompt, Tobias lets an AI agent take it from there. The agent builds the application, registers it with the JTL Cloud Platform, connects it to ERP data, and deploys it online.The entire process takes about 20 minutes, even with an internet outage along the way.What makes this interesting goes beyond the time saved. The agent uses the JTL Developer Docs to figure out how to work with the platform, then refines the application as Tobias gives it more instructions.It’s a practical look at what building a JTL Cloud App with an AI agent looks like today, from the first prompt to a working application.

What the Agent Builds

The task is to build a customer dashboard that answers two questions: which customers are most valuable, and what has a particular customer purchased?Tobias gives the agent these requirements. It builds two views:
  • A customer dashboard. A standalone app for viewing customer information, accessible both on its own and through the third-party applications section of Cloud ERP.
  • A customer Panel. A contextual view that opens alongside a selected customer record, showing their purchase history and payment status.
The two views serve different purposes. The dashboard helps you compare customers at a glance. The Panel brings purchase history into the workflow when you’re already looking at an individual customer.Building both is straightforward to describe. The interesting part is getting an AI agent to build them against the JTL Cloud Platform, connect to real ERP data, and follow the conventions that make the application work within the existing ERP environment.

Give the Agent Access to the JTL Developer Docs

An AI agent can generate code, but it needs to understand the platform it’s building for. Without the right context, it may make incorrect assumptions about how JTL apps are structured, authenticated, or registered.The JTL Developer Docs MCP server gives the agent access to the developer documentation while it works. MCP (Model Context Protocol) provides a standard way for AI applications to access external tools and information, allowing the agent to look up relevant JTL guides instead of relying on its existing knowledge alone.The JTL Developer Docs include a Build with AI guide that explains how to connect the docs MCP server to supported agents.

Set up the Agent to Build, Not Just Chat

Access to the documentation is only part of the setup. The agent also needs to work with your project: create files, install dependencies, run commands, and register the app with the platform.Tobias uses ChatGPT with Codex enabled. Other agents can follow a similar workflow, provided they support the JTL Developer Docs MCP server and can work with files and commands on your machine.You’ll also need a JTL account, an organization, and an On-Prem ERP connected to the JTL Cloud Platform. In the walkthrough, this initial platform setup takes around 10 to 15 minutes.With the documentation connected, the necessary tools available, and the platform ready, you can move on to building the app.

Describe the Outcome You Want

Tobias starts with a straightforward prompt:
Please use the JTL Docs to build and register an app. I want the app to show me an overview of my most valuable customers. When I select a customer, I want to see an overview of their purchase history. Please also add a Panel that shows this history when I select a customer.
The prompt describes what the app should do, without prescribing how to build it. It leaves the agent to choose the implementation, using the JTL Developer Docs to understand the platform and its requirements.The agent starts querying the docs, creating the project, and running the commands needed to build and register the app. When it asks which organization to use, Tobias selects his test organization and lets it continue.

Registration and Installation are Different

Registering an app makes it available on the JTL Cloud Platform. Installing it connects the app to a specific organization and requires approval for the access it requests.After registration, the agent opens the installation dialog. Tobias reviews the requested access and approves the installation for his organization.The agent then opens the standalone app. The dashboard displays data from Tobias’s On-Prem ERP using his existing JTL Cloud sign-in.The app is also available from the third-party applications section of JTL Cloud ERP. From there, Tobias can open a customer record and view the second part of the build: a Panel showing that customer’s purchase history.Both views are now working as requested, with the standalone dashboard providing the overview and the Panel bringing purchase history into the ERP workflow.Customer value dashboard ranking customers by gross order value, with totals for order value, purchasing customers, and orders

Improve the App Through Follow-up Prompts

The first version doesn’t have to be the final one.While exploring the app, Tobias notices that the item names in the customer Panel aren’t clickable. He asks the agent:
Can we make the items in the Panel clickable links?
The agent updates the app so the item names open the relevant items in a new tab.This is how Tobias continues working with the agent: build a first version, try it out, and ask for specific improvements. He can keep refining the app through prompts or open the generated project in his editor and take over when he needs more control.Customer Panel beside the customer list in Cloud ERP, showing a customer's purchase history with linked order numbers, payment status, and delivery status

Deploy the App so Your Team Can Use it

So far, the app runs on Tobias’s machine. To make it available to others in his organization, he asks the agent to deploy it:
Please deploy my application so it is reachable by other people.
The agent handles the deployment and updates the app’s registered URL on the JTL Cloud Platform. A few minutes later, Tobias opens the hosted version and confirms that it loads.He also tests the standalone URL in a private browser window. Without an authenticated session, the browser redirects to JTL login.In this demonstration, Tobias uses the hosting integrated into ChatGPT. Other hosting options are possible, depending on your tooling and requirements.From the initial prompt to a deployed app, the walkthrough takes about 20 minutes.

A Working App is the Starting Point

Building a working application in 20 minutes is a useful result, but it doesn’t make the application production-ready.Before relying on it in daily operations, review the permissions it requests and have a developer inspect the generated code, particularly its authentication, authorization, error handling, and treatment of sensitive data. Test failure scenarios as well as the happy path, including empty records, unexpected data, and failed requests. You’ll also need a plan for maintaining the application and its hosting as requirements change.The JTL login flow shown in the demonstration is a useful part of the setup, but a login redirect alone doesn’t establish that the application’s security and access controls are correct.AI can accelerate the initial implementation. Deciding whether the result is secure, reliable, and ready to deploy still requires engineering judgment.

Try it With Your Own Idea

This walkthrough is the first part of a series on building JTL Cloud Apps with AI. Upcoming videos will cover sharing an app with other businesses and publishing one to the JTL App Store.To get started, visit the JTL Developer Docs. The Build with AI guide covers connecting to the docs MCP server, and the account setup guide covers the account and Wawi setup needed for the walkthrough.Start with a problem you understand well: information your team repeatedly looks up, a customer workflow that takes too many steps, or an overview you’d like to see directly in Cloud ERP.Describe the outcome you want, give your agent access to the JTL Developer Docs, and build from there.If you get stuck, the Help section of the developer docs links to the JTL Discord community, where you can ask questions and get guidance from other developers and JTL engineers.