Skip to main content
4TrustConsulting GmbH
← Back to SAP topics

13 November 2026

AI agents on the ABAP platform: changes to the ERP core need guardrails

The head of ABAP Platform at SAP describes ABAP as the foundation on which AI agents are to carry out real business transactions, and names authorisation scope, audit trail and isolated test environments as prerequisites. For programme leadership, this is an occasion to check your own change and approval path for agents.

What happened

In an interview in the SAP News Center on 11 November 2026, the head of ABAP Platform at SAP describes how the platform is to become the foundation on which AI agents carry out real business transactions, and which guardrails she considers necessary for this.

Who is affected

The interview addresses large enterprises whose business processes run on the ABAP platform, and with them IT management, development, operations and the programme leadership of a transformation. According to the interview, agents are to sit in front of the applications as an additional layer; it names financial postings, HR processes and supply chain decisions as examples. As requirements for the platform, the interview names reliable, traceable and reproducible results, a clearly defined authorisation scope for every agent with an audit trail of who triggered it and what it changed, and enough capacity for peaks in agent activity. In the interviewee's words, agents for administration and operations are to have a supporting role only and not act autonomously in the system. For migrations, SAP is building seven agent families according to the interview, covering system analysis, testing and custom code migration among others, which are also meant to be usable on their own for custom code quality reviews. ABAP AI capabilities are now also available in earlier releases, it says; the interview does not say which releases or how they are licensed.

What to do

For programme leadership, a side remark is decisive: according to the interview, changes in ABAP take effect immediately in the system and are generally corrected forward, not rolled back. We would therefore stipulate that every change an agent proposes or makes takes the same path as any other – through development and test systems, with code review and business acceptance, before a transport reaches production. In our view, agents should appear in the authorisation concept as technical users of their own with their own role design, rather than working with the rights of the person who triggers them. And the agents' log should fit the existing documentation for internal and external audit. As a starting point, we would consider the code quality review, where an agent assesses but changes nothing.

Our assessment

The interview speaks from the vendor's perspective and gives no figures on effort saved or on the quality of the results. We find it notable how clearly it leaves responsibility with people: developers' tasks are changing, "the accountability isn't". From large SAP programmes we know testing and cut-over as the phases that reveal whether business, IT and implementation partner mean the same thing by "done". If agents produce more changes, the bottleneck shifts, in our view, from writing code to testing and acceptance. A programme should plan capacity for this before it lets agents into the ERP core.

Sources

Assessment by 4Trust as of the date stated; the occasion is reproduced as stated in the linked source above. This article is a general assessment and does not replace advice on an individual case. If you spot an error, please let us know: info@4trust-consulting.com.