When Anthropic announced Claude Managed Agents, most people were still reading the press release. Three hours later, Lemongrass Solution Architect Sam Stuttard had already wired the new agent runtime into SAP via SAP Business Technology Platform (BTP) and was watching an autonomous AI agent create sales orders and deliveries inside an SAP system.
It’s the kind of hands‑on experimentation that shows where the SAP ecosystem is heading — and how quickly.
What Sam built wasn’t a mock‑up or a theoretical architecture diagram. It was a working prototype that demonstrated, in real time, how an agentic AI model can read SAP data, analyze it, make decisions, and post transactions back into SAP, all while staying within the governance and security boundaries of BTP. And given how new the technology is, he may well have been one of the first people globally to try it.
A New Kind of AI Arrives
Claude Managed Agents introduce a different way of working with AI. Instead of prompting a model step by step, you give the agent a goal and a set of tools, and it handles the orchestration itself. It runs inside its own container, chooses which tools to use, and can even spawn sub‑agents to break down complex tasks.
For SAP customers, the obvious question is whether this kind of autonomy can be used safely with enterprise systems. Sam set out to answer that by connecting the agent runtime to SAP through BTP — the layer that provides the governance, security, and control SAP landscapes demand.
The agent itself runs in Anthropic’s cloud, but every interaction with SAP flows through BTP. That means SAP authorizations are respected, audit trails remain intact, and Cloud Connector can be used for private cloud or on‑prem systems. In other words, the AI can think freely, but it can only act within the boundaries you define.
What the Agent Actually Did
To show how the agent reasons, Sam broke the workflow into phases. The agent began by reviewing upcoming contract activity and identifying which scheduled lines were due in the next week. It then analyzed the associated business partners, checking master data, historical behavior, and credit‑related information. When it spotted data quality issues, it flagged them for human review rather than pushing ahead blindly.
Next, it moved on to product data, gathering the information it needed to create orders. Based on everything it had learned, the agent made a judgement call: it skipped one potential order because it wasn’t satisfied with the customer’s payment history. The remaining three orders were created automatically in the SAP system — not simulated, not mocked — and the agent then generated the corresponding delivery documents.
All of this happened inside a container the agent spun up for itself, and the entire process took less than three minutes. Sam streamed the events back into a BTP application so he could watch the agent’s decisions unfold in real time, complete with an executive summary of what it had done and why.
This wasn’t a chatbot answering questions. It was an autonomous system executing SAP business processes.
Why This Matters for SAP Customers
What Sam demonstrated is that agentic AI can interact with SAP in a way that is both powerful and safe. BTP acts as the control plane, exposing only the tools and APIs the agent is allowed to use. The agent can think as creatively as it wants, but it can only act within the guardrails you define.
That opens the door to a new class of automation. Sam’s demo focused on sales orders and deliveries, but the same pattern could apply to invoice reconciliation, production scheduling, or even multi‑step processes like financial month‑end close. In fact, he’s already experimenting with an eight‑step month‑end accelerator that breaks the workload into smaller tasks and lets the agent spawn sub‑agents to handle them in parallel.
What’s striking is how little friction there was in getting the prototype running. With BTP, the Cloud SDK, and an API key, Sam was able to build something meaningful in an afternoon. If that’s possible in a sandbox system with demo data, imagine what could be achieved in a quality or pre‑production environment with real business data.
A Glimpse of What’s Coming
This demo hints at a future where AI doesn’t just assist SAP users — it actively runs SAP processes. Not by bypassing governance or introducing risk, but by working through the same secure, auditable pathways organizations already trust.
The technology is still new, but the direction is clear. Agentic AI will accelerate automation far beyond what RPA or workflow tools can offer, and BTP will be the layer that makes it safe.
For SAP leaders, this is the moment to start experimenting. The organizations that build familiarity now will be the ones ready to take advantage of what’s coming next.
