ABAP Platform Becoming an Enterprise-Grade Execution Backbone for Agentic AI

In an interview, Sonja Liénard, head of ABAP Platform at SAP, discusses the Autonomous Enterprise and ABAP platform as an “execution backbone” for agentic AI.

What is the Autonomous Enterprise? What part does ABAP platform play in it?

Liénard: At SAP Sapphire in 2026, we unveiled the Autonomous Enterprise—our vision for how we want to help large enterprises in the era of agentic AI. The fundamental shift we’re seeing is in the way people work with technology. Interfaces are no longer the focus: interaction is outcome-driven, not tool-driven.

The user experience is being transformed. Instead of filling out forms in an application, you simply open a chat and ask: “Who was on vacation in the last 30 days, and when?” The answer comes back in real time. That’s the Autonomous Enterprise.

The nature and purpose of ABAP platform is not changing. It is still the foundation on which business processes run—and it is now also the backbone on which agents operate. What is changing is the way users access applications. Agents now sit in front of those applications as an additional abstraction layer.

If agents are expected to handle real business tasks—not demo scenarios, but actual financial postings, HR processes, or supply chain decisions—they need to run on a platform that is reliable, governed, and scalable.

Can you explain what ”outcome-driven” means?

Here’s an example. Instead of filling out a requisition form field by field, I open my chat and say, “I need 30 laptops for new hires by the end of October.” Agents check the catalog, review the budget, and create the requisition, without me having to navigate the process myself.

As a user, I’m focused solely on the outcome, not on the route I need to take. When I enter a natural-language prompt, agents can autonomously pursue a variety of solution paths. What we’re seeing is a paradigm shift. The form-based environment we’re used to is giving way to the outcome-driven, agentic model.

What does designing a platform for agents involve?

It presents us with challenges on various levels.

The first is reliability. Data and results must always be reliable, traceable, and reproducible—regardless of who triggered the agent or which path it took to obtain the data. This requires safety nets and very robust quality controls and tools.

The next level is accountability and governance. Each agent must have a clearly defined authorization scope. What is it permitted to do? What is it not permitted to do? What data is it allowed to access? Agents need skills and workflows to follow. And there must also be an audit trail: what changes an agent made, who triggered the agent, and what exactly it was instructed to do. Well-defined interfaces are essential to ensure that agents operate strictly within their architectural boundaries, helping to prevent unauthorized data flows, close potential security gaps, and protect systems from compromise.

And finally, there’s scalability. ABAP platform must have the computing and storage capacity required to handle spikes in agent activity. We are preparing for this, running security tests and assessing the platform’s scalability limits.

The key principle remains the same: we must define very clearly where and when humans retain the final say. That is the foundation of enterprise trust—in the agentic AI era as much as in any other.

Previously, you described the ABAP MCP (Model Context Protocol) server as an important building block for everything related to agentic AI. What’s the latest news here?

We released the ABAP MCP server in the second quarter. It opens ABAP platform to the agentic era by enabling MCP-compatible agents to interact with ABAP capabilities. MCP can provide well-defined interfaces for agent-to-agent communication, interaction with external tools, and task delegation.

You’re referring to agents that perform tasks in ABAP systems. But what about agents that administer or operate these systems?

We released our first AI-powered development tools in early 2025. They were designed to boost developer productivity by providing assistance with coding and with code quality reviews. We’ve now gone further. The market is moving fast, and the next logical step is to extend AI capabilities to operations.

We’re now looking at agents for administration and operations—and investing there, too. But we need safety nets, clear accountability, and well-defined boundaries. We will work to ensure that these agents have a supporting capacity only and that they do not operate in systems autonomously.

With agents handling more and more operational tasks, how will SAP ensure that the guardrails remain sufficiently robust?

In regards to development, it’s important to understand that ABAP and non-ABAP development differ. In ABAP, changes take effect immediately in the system and are generally corrected forward, not rolled back. This makes it especially important to have robust guardrails in place for agents, such as an isolated testing environment, thorough quality checks, and clearly defined skills and playbooks that govern agent actions and ensure reproducible results. We’re investing in all of this.

What about customers that are not yet on the latest release?

ABAP AI capabilities are now available in earlier versions as well. For enterprises looking to move to a newer release, our agent-led toolchain can offer dedicated support. We are currently building seven agent families that cover different process steps of a migration, including system analysis, testing, and custom code migration. These agents can also be deployed independently for custom code quality reviews.

With an eye to the rest of 2026, what should customers and partners look out for?

SAP is focusing on three elements. The first is migration. By year-end, we plan to ship additional practical tools to help support the entire migration process. Second is sovereign cloud solutions, for which we are developing compelling AI capabilities for enterprises with specific data sovereignty requirements. And third is MCP tooling, well-defined, secure external interfaces that open ABAP platform to the agentic world without compromising platform security.

You’ve been clear throughout that ABAP developers will stay relevant and that AI will not replace them. But the picture you’re painting is of a platform that increasingly runs itself. How should developers be thinking about their role?

How is the developer role changing? That’s a question for all developers, not just those who work on ABAP platform. Humans will continue to be responsible for what enters the system and what happens to data. We need to define clearly where humans make the final call and who is accountable. So, the role of developers is shifting. They are spending less time on writing code and more on oversight, software architecture, and prompting to ensure that agents do exactly what they are supposed to. The tasks are changing; the accountability isn’t.

And there’s one more point I’d like to stress: ABAP platform is becoming an enterprise-grade execution backbone for agentic AI. I use the word “backbone” deliberately. We’re not a generic agent runtime. General-purpose AI can’t match what we offer: rich, proven business logic, enterprise-grade governance, and a 40-year track record of running the world’s most critical business processes. That’s the role of ABAP platform in the Autonomous Enterprise.