Clean Core in the Agentic Era: Why SAP's A–D Clean Core Levels Decide Your AI Readiness
For CIOs, SAP CoE leads and transformation partners planning RISE with SAP, S/4HANA Cloud Private Edition or Business Suite journeys alongside AI agents.
Clean core used to be sold as an upgrade-cost story. In 2026 it is an AI story. AI agents — whether Joule Agents, Joule Studio agents or pro-code agents on SAP BTP — act through APIs, events and business objects. If those interfaces are undocumented, modified or unstable, agents become unreliable after every upgrade. A clean core is what makes agents trustworthy.
1. The clean core level concept in one minute
SAP evolved its earlier three-tier extensibility model into the clean core level concept, a maturity model with four levels:
Level A — extensions using only released, stable public interfaces: ABAP Cloud, key-user extensibility, and side-by-side extensions on SAP BTP (SAP Build, CAP, Joule Studio). The gold standard and default for all new development.
Level B — classic ABAP using classic APIs and technologies following best practices. Considered clean core and generally upgrade-stable.
Level C — use of internal SAP objects. "Conditionally clean": acceptable with governance, change-log tracking and planned remediation.
Level D — modifications and practices SAP does not recommend. Not clean; highest upgrade risk.
An important nuance: an extension takes the level of its least clean component. One unreleased call can drop an otherwise Level A extension.
2. Why agents raise the stakes
Agents need stable tools. MCP tools and A2A skills should sit on released APIs and CDS views. A tool built on an internal table read silently breaks when SAP changes it.
Agents need semantics. Released CDS views and business objects carry meaning (field labels, associations, authorisation). That metadata is what lets an LLM reason correctly.
Agents need authorisation that holds. Released APIs respect the standard authorisation model; direct table access often doesn't.
SAP's own agents assume standard processes. Heavily modified processes reduce what Joule Agents can do out of the box.
3. A practical clean-core roadmap for AI
Assess. Run ATC with the clean-core checks and the Custom Code Migration app; classify every custom object A–D and identify unused code.
Retire. Delete unused custom code first — often the biggest, cheapest win.
Remediate D, then C. Replace modifications with key-user extensibility, released BAdIs, ABAP Cloud (on-stack) or BTP side-by-side extensions.
Move logic side-by-side where it differentiates. Build process extensions with CAP on BTP, Fiori elements UIs, and events via SAP Event Mesh / Advanced Event Mesh.
Expose agent-ready tools. Publish curated APIs through the Integration Suite MCP gateway or CAP @mcp services — only on Level A/B foundations.
Govern continuously. Track the percentage of code per level as a KPI in your CoE dashboard and gate new developments at Level A.
4. Where BTP extensions fit
Side-by-side extensions are central to clean core. Typical patterns we deliver:
Approval and exception workflows in SAP Build Process Automation, triggered by S/4HANA business events.
CAP-based apps that combine S/4HANA, Ariba and SuccessFactors data via remote services, with Fiori elements UIs.
AI agents on the same CAP services — the extension and the agent share one governed tool layer.
5. Common objections — answered
"Our Level C code works fine." It works until the next upgrade. Level C is allowed with governance, but it should be visible and on a roadmap.
"Side-by-side is slower." Not with CAP and modern tooling. MCP-aware development tools for CAP, UI5 and Fiori have significantly shortened delivery cycles.
"We'll clean up after go-live." Every agent you build on unclean interfaces adds remediation debt. Clean the interfaces your first agents depend on before you build them.
FAQ
Does clean core apply to private cloud? Yes — the level concept was designed precisely for SAP Cloud ERP Private and on-premise systems with existing custom code.
Is classic ABAP banned? No. Well-written classic ABAP using classic APIs can be Level B and clean core.
Get a Clean Core + AI Readiness Assessment
Raven Infotech combines custom-code analysis with an AI use-case map: you get your A–D distribution, a remediation backlog and the first three agent-ready tools to build. Book a free assessment call.
Comments