I’ve been working in SAP for close to 20 years now. Over that time, I’ve seen the development landscape evolve from SE38 and SE80 to Eclipse ADT, from classical ABAP to ABAP OO, and now RAP. But nothing has shifted the day-to-day developer experience as quickly as AI-assisted coding.
When our team started discussing AI tools for ABAP development, the first name that came up was SAP Joule. It looked promising, deeply trained on SAP code, integrated into Eclipse ADT, and purpose-built for ABAP. But when I dug into the availability details, I hit a wall: Joule for Developers is not available for on-premise S/4HANA systems. Our landscape is on-premises, and that wasn’t going to change overnight.
So I started exploring what else was out there. I looked at Amazon Q Developer, which has strong ABAP capabilities and an impressive ABAP Accelerator MCP server. But it required standing up AWS infrastructure. Then I checked GitHub Copilot and realized our organization already had Copilot Business licenses under our GitHub Enterprise agreement. The path of least resistance was clear.
I set it up on a Friday afternoon. By Monday morning, I was using it in real development work. The speed at which I became productive surprised me. I honestly expected a longer ramp-up.
I initially set up Copilot in both Eclipse ADT and VS Code, expecting roughly the same experience. I was wrong. The Eclipse plugin gives you solid code completions and a useful chat panel, but VS Code with Agent Mode is on a different level entirely. The first time I described a multi-step task in natural language and watched the agent plan its approach, create files, and iterate on errors autonomously, I knew this was going to change how my team works.
One moment that stood out: I pointed Agent Mode at a legacy report that nobody on the current team had written. Within minutes, it had analyzed the entire program, generated a functional description, flagged two performance issues with nested SELECTs, and suggested refactored code with JOINs. That single interaction would have taken a developer half a day to do manually.
Another area where I’ve seen real value is unit test generation. Let’s be honest, ABAP Unit tests are often the first thing that gets cut from project timelines. With Copilot’s /tests command, you can generate a solid test scaffold in seconds. It’s not perfect, you still need to review and adjust, but it removes the biggest barrier, which is the blank-page problem of writing test classes from scratch.
If you’re on an on-premise SAP system, here’s where things stand today:
Solution | On-Premise? | IDE Support | Key Consideration |
SAP Joule | No | Eclipse ADT | Cloud only (BTP / S/4HANA Cloud) |
Amazon Q Developer | Yes | Eclipse, VS Code, BAS | Requires AWS ecosystem |
GitHub Copilot | Yes | Eclipse ADT, VS Code | Broad model choice, mature Agent Mode |
I went with GitHub Copilot because we already had the licenses. Your mileage may vary, if your organization is on AWS, Amazon Q Developer is absolutely worth evaluating. The important thing is that on-premise ABAP teams are no longer locked out of AI-assisted development.
For AI-powered development tasks, VS Code with Copilot Agent Mode is in a different league.
Eclipse ADT + Copilot: Good code completions, a helpful chat panel, and basic agent support that’s been improving with each release. For day-to-day ABAP work, it’s a meaningful step forward.
VS Code + Copilot: Everything Eclipse has, plus fully autonomous Agent Mode with multi-step planning, terminal integration, and direct SAP system awareness through the ABAP FS extension. This is where I do my complex development work now.
Here’s the exact setup I followed. The whole process took me about 30–45 minutes, and I was productive immediately after.
Step 1: Install Visual Studio Code
Download and install the latest version of VS Code from code.visualstudio.com. I’d recommend getting the latest stable release — older versions may not support all Copilot features.
Step 2: Install ABAP Remote Filesystem Extension
This is the extension that makes VS Code talk to your SAP system. Without it, Copilot has no ABAP context — with it, the AI can see your actual system objects.
Add the SAP system connection details. I configured both our DEV and SBX systems here. Use placeholder values if you’re just testing the setup initially.
Step 3: Install SAP Web IDE Adapter for VS Code
This extension enables SAP-specific development workflows within VS Code. It’s a small addition but it rounds out the SAP development experience.
Step 4: Install SAP Joule Studio Code Extension
Even though Joule isn’t available for on-premise ABAP, I installed the Joule Studio Code Editor extension anyway — it’s useful if you also work with BTP environments, and it doesn’t interfere with Copilot.
Step 5: Install GitHub Copilot Extension
This is the main event. Install the GitHub Copilot extension and sign in with your company’s GitHub account.
A tip from my experience: if your company uses SSO for GitHub, make sure you complete the authentication through the browser window that pops up. I initially missed this and spent 10 minutes wondering why Copilot wasn’t activating.
Step 6: Activate ADT Service on SAP System
This is the SAP Basis step. Make sure the ADT service is active in transaction SICF. If your Basis team hasn’t already enabled this for Eclipse ADT, you’ll need to request it.
Step 7: Connect to the SAP System
Now connect VS Code to your SAP system. This is the moment where everything comes together.
Step 8: Verify SAP System Connection
Once connected, you should see your SAP system’s object tree in the Explorer pane. When I first saw our $TMP package, Core Data Services, and Source Code Library appear in VS Code, it felt like two worlds merging.
Step 9: Verify GitHub Copilot and Model Availability
Finally, confirm that Copilot is active and check which AI models your organization has approved. In our case, we have access to Claude Haiku, Claude Opus, Claude Sonnet, and GPT models. I primarily use Claude Haiku for quick tasks and Claude Sonnet for complex agent work.
Once installed, open the Copilot Chat panel (Ctrl+Alt+I) and choose your mode. Here’s how I use each one in practice:
I use this multiple times a day. Instead of searching SAP Help or opening a browser, I just ask Copilot directly in the IDE. It’s especially useful for BAPIs and function modules I don’t work with regularly.
“Explain how BAPI_USER_GET_DETAIL works”
“Best practices for exception handling in ABAP OO?”
When I have existing code open and want specific changes, I select the code, press Ctrl+I, and describe what I need. This has been particularly useful for modernizing older code patterns.
“Refactor this to use NEW instead of CREATE OBJECT”
“Optimize this to use a JOIN instead of nested SELECTs”
This is the feature that changed my workflow. I describe a complex task and let the agent figure out the approach. It plans, generates, reviews its own output, and iterates. Not every result is perfect on the first try, but the starting point it gives you is usually 70–80% there.
“Create an ABAP class ZCL_BP_VALIDATOR with unit tests for BP number validation”
“Analyze program ZSALES_ORDER_REPORT — generate documentation and flag performance issues”
“Create a complete RAP application for insurance claims with CDS views, behavior definition, and service binding”
These are quick shortcuts in the chat panel that I’ve made part of my routine:
One thing I didn’t expect was how smooth object creation is directly from VS Code. Enter the ‘>’ command to list options, then select ABAP Create Object.
For searching existing objects, use the ABAP Search for Object command. The filter option is handy when you’re looking for a specific object type.
If you set up only one additional thing beyond the basic installation, make it this. Custom Instructions tell Copilot how your team writes code — your naming conventions, your preferred patterns, your performance rules. Without them, Copilot gives you generic ABAP. With them, it gives you code that looks like your team wrote it.
Create a file called .github/copilot-instructions.md under each connected system. You define your rules in plain English, and Copilot follows them automatically across all suggestions.
The instructions I set up for our team enforce:
A personal observation: after deploying Custom Instructions, I noticed the quality of Copilot’s ABAP suggestions improved dramatically. It stopped suggesting CREATE OBJECT and started using NEW. It stopped generating SELECT * and started using explicit field lists. These are small things individually, but across a team of developers, they add up to consistently better code.
Here are the five ways GitHub Copilot has become part of our regular workflow:
I want to be transparent about where Copilot falls short for ABAP, because I think setting realistic expectations is important:
AI-assisted ABAP development works on-premise today. GitHub Copilot in VS Code, especially Agent Mode, has genuinely changed how I work and how my team approaches development tasks. The setup takes an afternoon, and the productivity gains start from day one.
If your organization has GitHub Copilot licenses and an on-premise SAP landscape, I’d encourage you to try this setup.
I’d love to hear from others in the community who are exploring AI tools for ABAP development. What’s working for you? What challenges have you faced? Let’s keep the conversation going.
#SAP #ABAP #GitHubCopilot #VSCode #AgentMode #AI #OnPremise #S4HANA #GenAI
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
| User | Count |
|---|---|
| 25 | |
| 19 | |
| 17 | |
| 13 | |
| 11 | |
| 11 | |
| 10 | |
| 9 | |
| 6 | |
| 4 |