Updated 10 September 2026. This article replaces the earlier career-market overview with practical learning priorities for ABAP developers.
Working effectively with AI on an SAP project requires more than prompting. A developer needs to understand the system context, recognise an incorrect proposal and produce evidence that a change meets its requirements. These are useful skills to practise whether you work on SAP ECC or on-premise S/4HANA.
1. Explain an existing program from its sources
Choose a small custom report in an authorised development system. Identify its inputs, outputs, processing steps, dictionary dependencies and error paths. Separate facts visible in the source from assumptions about business use. Ask an AI assistant for an explanation, then check its claims against the objects it actually inspected.
Deliverable: a short program overview with source references and a list of unresolved questions.
2. Turn a request into testable behaviour
Write a before-and-after example before asking for code. Include normal input, boundary values and invalid input. Decide which checks belong in unit tests and which require integration testing. A validator method and a complete report execution provide different coverage.
Deliverable: acceptance criteria linked to test scenarios, including the expected results.
3. Understand the connection between the AI client and SAP
A model, an AI client and an MCP server are different components. The client handles interaction with the model; an MCP server exposes configured tools. An available tool does not prove that the SAP user has the right permissions or that an operation works on the installed backend.
- AI client: loads project instructions and requests tools.
- MCP connector: exposes the configured operations.
- SAP backend: executes permitted operations on the deployed system.
- Developer review: checks source, execution results and unresolved assumptions.
Review the model-provider and storage data paths separately; a local connector does not mean all data stays local.
For a concrete example, the ABAPilot client setup guide distinguishes Claude Code, Gemini CLI and GitHub Copilot in VS Code. Configuration formats differ. ABAPilot by Crimson Consulting uses a public connector and a separately licensed SAP backend; installing the connector alone is not a complete setup.
4. Review generated code and executed checks
Inspect the proposed diff for behaviour changes outside the request. Check dictionary references against the system. Record syntax, activation and test results separately. A proposed test is not an executed test, and passing unit tests do not establish every integration path.
Deliverable: a reviewed diff and an execution record that identifies checks not run.
5. Keep documentation aligned with the implementation
Update the technical specification after the final change. Link requirements to the implemented behaviour and test evidence. Record limitations where the next developer will find them, rather than leaving them only in a chat transcript.
In our recorded report-change example, the final sandbox run passed 14 ABAP Unit methods and updated both the technical specification and test scenarios. Manual report integration was not executed. That distinction is part of the handover.
A practice exercise for your team
- Select an existing custom program and an agreed sandbox scope.
- Document its behaviour and dependencies without changing it.
- Define one small change and its acceptance examples.
- Review the implementation, run relevant checks and record actual results.
- Ask another developer to review the specification and remaining coverage gaps.
Use synthetic data where possible and follow your organisation’s access and data-handling rules. Compare the complete effort, including corrections and review, with your normal workflow. Do not treat a polished AI response as evidence of time saved.
Choose learning material around your current work
Start with the tasks your team maintains: reports, interfaces, enhancements or applications. Learn the supported development approach for your actual SAP release. A tool demonstrated on one landscape does not establish compatibility elsewhere.
Continue with the AI for ABAP development evaluation guide, the SAP ECC guide or our ABAP change-delivery checklist. These articles are written by Crimson Consulting, the vendor of ABAPilot; they are practical guidance, not independent product rankings or salary forecasts.