SAP Rewrites Custom Code with AI Agents, Previewed at TechEd Berlin

Title

SAP Rewrites Custom Code with AI Agents, Previewed at TechEd Berlin

Meta Description

SAP TechEd Berlin focuses on AI Agent-driven custom code migration. This article outlines the SAP agenda, the four Clean Core levels framework, scenarios where Agents excel and pose risks, plus an on-site question checklist and post-migration self-assessment methods.

Article Content

SAP is betting on a bold proposition: the ECC custom code migration challenge that has plagued countless enterprises for years can be taken over by AI Agents. TechEd Berlin will be held from October 27 to 29, with virtual attendance free. SAP has scheduled multiple sessions on Agent-driven custom code migration, but between the agenda and verifiable results lies a whole set of evaluation criteria that need serious scrutiny.

What SAP Actually Put on the Agenda

Every ECC migration has a common "villain." It's not the budget, not the database, but custom code—tens of thousands of Z programs, user exits, forms, and reports, written over twenty years, understood by only three people, two of whom have already left.

According to SAP's official TechEd 2026 ABAP agenda page, the following sessions are worth noting:

• TM105, Agent-Driven Custom Code Migration: From Planning to Large-Scale Remediation. SAP's description states that custom code migration is often the bottleneck in SAP ECC to SAP Cloud ERP Private migrations. The toolchain extends from planning in SAP Cloud ALM to "AI-driven large-scale remediation," with the core component being the Custom Code Migration Agent in SAP Joule for Developers. There is one session each on October 27 and 29.
• TM260, Hands-On Workshop: Using MCP tools to complete custom code migration and modernization from ECC to RISE with Agentic AI.
• TM103, Reimagine Migration and Modernization with Agent-Driven AI Assistants. SAP's exact words: reduce costs, accelerate timelines, and cover the full migration lifecycle.
• TM108, The Clean Core Journey for SAP Cloud ERP Private, including how AI helps custom code achieve Clean Core compliance and new guidance related to SAP API strategy.
• AD100 and AD819, Agent-Driven ABAP Development, based on ABAP Language Server and ABAP MCP Server. SAP's tone for the entire track: development is moving from "AI-assisted" to "Agent-driven."

A signal can be read from the agenda itself: SAP is no longer just treating AI as a code completion tool, but trying to place it into the decision-making and execution stages of the migration process.

First, Understand the Framework: The Four Clean Core Levels

To judge whether an Agent has truly "fixed" code, you must first know SAP's definition of "good." SAP introduced Clean Core Levels in August 2025, replacing the previous three-tier model:

• Level A: Uses only publicly released, stable interfaces backed by formal stability contracts. Built on BTP or in-stack via ABAP Cloud.
• Level B: Also uses SAP classic APIs and technologies, documented and usually upgrade-stable.
• Level C: Accesses SAP internal objects. Flexible for legacy scenarios but carries upgrade risk. SAP plans to introduce a change log for SAP objects to detect changes early.
• Level D: Not recommended. Modifications, write access to SAP tables, implicit enhancements. SAP calls this the highest risk and a source of major technical debt.

A notable detail: SAP's own community guidance acknowledges that the old binary view—"Clean Core" versus "not Clean"—is "too strict" for systems with large amounts of classic custom code. In other words, SAP revised the model precisely because real environments could not meet the old standard.

On the governance side, SAP recommends ABAP Test Cockpit and states that central ATC on SAP BTP is its recommended solution for Clean Core development governance in private cloud and on-premise environments.

ABAP Test Cockpit (ATC) is a static code analysis tool provided by SAP for checking the quality, security, and compliance of ABAP code. It can scan custom code and identify objects that do not meet Clean Core requirements, such as direct access to SAP internal tables, implicit enhancements, or modifications. ATC supports running during development, testing, and operations phases and can be integrated into CI/CD pipelines. In Clean Core governance, ATC helps teams quantify the "cleanliness" of custom code and track improvement progress. SAP recommends using central ATC on BTP in private cloud and on-premise environments to achieve unified code governance across systems.

ABAP Cloud is the programming model SAP provides for cloud-ready ABAP development. It enforces the use of publicly released and stable interfaces and restricts access to SAP internal objects. ABAP Cloud is one of the key paths to achieving Clean Core Level A, and developers can use it in the ABAP environment on BTP or in-stack in S/4HANA Cloud. Compared with classic ABAP, ABAP Cloud ensures that extension code remains stable during upgrades through syntax checks and runtime restrictions. For enterprises looking to modernize custom code and meet Clean Core requirements, ABAP Cloud provides a clear development paradigm.

Why This Is About More Than One Conference

Custom code is one of the main reasons customers cite for delaying migration. One partner CEO told CIO.com that many customers are "stuck on ECC because of Z code and industry customizations." The DSAG chairman also pointed to complex system environments and skills shortages.

The timeline is also applying pressure. SAP says mainstream maintenance for ECC ends in 2027, with extended maintenance until 2030. If Agents can safely reduce the custom code workload, the arithmetic of migration changes. But that "if" is the whole story.

What Agents Are Good At, and Where the Risks Lie

The following is my judgment, not SAP's official position.

Suitable for machines: Repetitive, rule-driven changes. Replacing obsolete statements or calls, adapting to known simplified data models, batch-fixing the same pattern across hundreds of objects.

Risky: Code whose intent exists only in someone's head. A user exit that quietly fixes a business rule. A report with side effects. Any scenario where "compiles and passes ATC" does not equal "still correct for the business."

An Agent can pass technical checks and still break business outcomes. That is why review and testing are as important as the remediation itself.

MCP (Model Context Protocol) is an open protocol designed to let AI models securely connect to external tools and data sources. In the SAP context, MCP allows AI Agents to access resources such as ABAP Language Server, code repositories, and testing tools, enabling code analysis, modification suggestions, and automated refactoring. MCP provides Agents with standardized "tool calling" capabilities, enabling them to understand the development environment and perform concrete operations rather than merely generating text. This explains why SAP's agenda emphasizes using MCP tools for custom code migration—it allows Agents to interact with the real development toolchain.

Six Questions to Ask on Site

1. What percentage of fixes require no manual editing? Ask for numbers, and for the size and type of the sample. A demo on clean code proves nothing.
2. How are changes reviewed, tested, and transported? Does the Agent propose suggestions, or write directly to the development system?
3. What happens to code that cannot reach Level A or B? Rewrite, retire, or keep outside the core?
4. How does it perform on my release and add-ons? The session description talks about Cloud ERP Private as the target. Also ask about S/4HANA on-premise.
5. What is the running cost of large-scale remediation? Ask about consumption billing and who bears failed attempts.
6. Who is liable when an Agent-made change causes an incident? You need that answer in writing.

Self-Assessment Method After Attending

Rather than being convinced by a demo, run a small trial yourself:

1. Select three groups of custom objects: one simple, one typical, one tricky.
2. First estimate the manual effort. Before the Agent touches anything.
3. Run the Agent, then measure: the amount of editing required, tests passed, defects found in QA, and hours spent.
4. Set the criteria before you start. For example: most fixes are accepted without editing, and there are no severe regressions. If you set the criteria after seeing the results, you're only fooling yourself.
5. Scale only the parts that pass the test. Keep the tricky group as an honest check.

The Criterion Is Not the Demo, but the Numbers

Do not judge this by demos. Demos use clean code.

Judge by numbers from messy systems: the proportion of fixes accepted without editing, tests passed, hours saved, incidents caused. If SAP can produce these numbers, it will be genuinely big news for every ECC enterprise worldwide.

After TechEd Berlin, what is truly worth tracking is not what was promised, but what was actually demonstrated.

One-line Summary

SAP bets on AI Agents to solve the ECC custom code migration challenge, but actual results must be judged by numbers, not demos.

Paragraph Summary

SAP has scheduled multiple sessions at TechEd Berlin on AI Agent-driven custom code migration, with the core tool being the Custom Code Migration Agent in Joule for Developers. The article points out that custom code is the main bottleneck in ECC migration, for which SAP introduced the new Clean Core Levels A-D framework and acknowledged that the old standard was too strict. While Agents are suitable for rule-driven batch remediation, code whose intent is embedded in individual experience still requires human review. The author suggests enterprises evaluate the real value of Agents through small trials and key metrics (such as the proportion of fixes requiring no manual editing), rather than trusting demos.