SAP is all-in on making it easier for its users to build custom agents. At Sapphire 2026 in Orlando on Tuesday, the company announced a new managed version of Joule Studio, its platform for building, deploying, and running AI agents and Joule skills to automate workflows across SAP and third-party systems.
The expansion delivers on much of what SAP signaled at SAP TechEd in Berlin last November, when SAP Business Technology Platform (BTP) president Michael Ameling told The New Stack the company would pursue an “open strategy” because, in his words, “which horse would you bet on?”
What’s new
At Sapphire 2026, SAP pushes this forward. Cursor joins the list of supported coding tools, and AutoGen and LlamaIndex join the supported agent frameworks. The Agent2Agent (A2A) protocol is now bidirectional, allowing third-party agents to natively call Joule Agents within enterprise processes. And SAP is also adding more of its own models into the mix.
At TechEd 2025, the company launched SAP-RPT-1, its first relational foundation model. Now it is introducing the SAP Domain Models, a family of SAP-aware foundation models that brings together the second-generation SAP-ABAP-2 coding model and new specialized variants for SAP S/4HANA and Ariba. Early access is open now, with general availability planned for the third quarter of 2026.
SAP is also offering all customers and partners 12 months of free design-time access to Joule Studio, available now through SAP’s Early Adopter Care program, with general availability planned for the third quarter of 2026.
Joule Studio originally launched at Sapphire 2025 as part of SAP Build, with the skill builder generally available in July of that year and the agent builder following in December. But the agents’ customers ran on customer-managed SAP BTP environments. Developers still had to provision accounts, wire destinations through Cloud Connector, and size the compute. Basically, the studio was managed, but the runtime was not.
A managed Joule Studio
As Ameling tells The New Stack, the new Joule Studio is an evolution of the existing one.
“It’s a new Joule Studio, but it comes with a completely different experience. It’s SAP managed: no configuration, no setup, whatever. It’s there out of the box,” he says.
Users can build their agents, deploy them, and gain built-in access to enterprise features such as audit logging and data privacy protections out of the box.
“It’s a new Joule Studio, but it comes with a completely different experience. It’s SAP managed: no configuration, no setup, whatever. It’s there out of the box.”
The runtime also handles agent memory. Agents get persistent long-term storage backed by HANA Cloud, so state, preferences, and other accumulated context survive across sessions without the developer having to build that infrastructure from scratch.
There’s a second, more strategic benefit to running the platform end-to-end, as Ameling noted. “It’s simpler for the end user. But in addition, what it means is we can infuse new technologies without the customer doing anything,” he says. “With customer-managed scenarios, that’s very hard. For example, there’s OpenClaw — can we use it in our environment and put security requirements on top? This is what we can do now.”
No enterprise in its right mind is letting OpenClaw loose in its environments right now, given all of the security issues around it, but the point stands: by managing the environment in this time of rapid development, SAP can move more quickly if it controls more of the stack itself.
Context
Ameling also notes — and this is something virtually every large enterprise vendor is focusing on right now — that Joule Studio is aware of a company’s entire context. These agents, after all, work far better when you can provide them with more specialized knowledge.
“It brings this entire context into the game,” Ameling says. “So you come up with a simple prompt: ‘I want to put an agent that improves my invoices.’ It knows your landscape due to LeanIX and the Knowledge Graph. It knows you have an S/4 system ’23, your SuccessFactors system, and whatever. So it retrieves the right APIs, the exact APIs which you need in order to feed your agent.”
The service then generates a structured requirements document, plus scaffolding and code grounded in a real enterprise context.
Ameling says, “Usually this takes several months with multiple people. Now you can do it in a few minutes.”
How much better is that compared to what a developer could already build with a general-purpose LLM and one of the pro-code agent frameworks, though? Ameling raised this question himself: “Can’t I do this with a simple LLM and pro-code tooling? Fair question, and we did the comparison. The outcome is very different because the context is very different. At least 50 percent more context, which makes a fundamental difference.”
The claim virtually every enterprise vendor is making right now is that those vendors who already own the structured representation of how a company runs, be that processes, data, or identities, start the agentic era with an advantage.
SAP’s overall answer here is the new Business AI Platform, which unifies the SAP Business Technology Platform, SAP Business Data Cloud, and the company’s Business AI offerings into one governed environment. At its center sits the SAP Knowledge Graph, which maps every entity, process, and relationship across a customer’s SAP landscape.
Sitting on top is a new family of SAP-aware foundation models the company is calling the SAP Domain Models. They’re trained on SAP code, data, metadata, business processes, and architecture knowledge. This lineup includes the second-generation SAP-ABAP-2 coding model and new specialized variants for SAP S/4HANA and Ariba.
Redesigning for agents
As SAP gets ready for more agents to work across its services, the company also had to rethink how it builds its own infrastructure. The APIs and services underneath Joule Studio, for example, had to be redesigned for agents rather than humans.
“You design every API, every service for agents, not for humans… Before, you had an API and a UI on top, designed for humans with lazy loading — first 10 items, then the next 10. That doesn’t work for LLMs.”
“You design every API, every service for agents, not for humans,” Ameling says. “That’s a fundamental difference in engineering excellence. Before, you had an API and a UI on top, designed for humans with lazy loading — first 10 items, then the next 10. That doesn’t work for LLMs. You need the full set.”
Open to more coding tools
Another major part of SAP’s openness bet is support for third-party coding tools.
The list of supported tools now includes Claude Code, Cursor, and VS Code. The supported agent frameworks originally included LangGraph and Crew.AI, with AutoGen and LlamaIndex now joining them.
Ameling says, “Today, maybe it’s Claude Code. Tomorrow it’s OpenCode. We need the flexibility to switch — same as we need to switch between models. You will never beat the speed of that.” The same logic also applies to the model layer: “Whatever model we take today, we will take a different model tomorrow.”
The workflow extensibility layer is now n8n. At TechEd, SAP and n8n announced a partnership that lets Joule Studio agents and n8n agents interoperate. At Sapphire, n8n is getting embedded directly in the Business AI Platform, running on BTP at both design time and runtime under SAP’s enterprise controls. Developers can wrap an agent in an n8n workflow for tasks such as human-in-the-loop approvals, task-center routing, or agent-to-agent handoffs without leaving SAP’s governance perimeter.
Governance and the shadow AI problem
Another reason customers, and especially their IT departments, might care about a managed runtime is that everything running on it inherits the platform’s governance.
Each agent at runtime receives a unique identity by default from SAP Cloud Identity Services. Customers who already have Okta or Microsoft Entra can plug those in as the identity provider, with SAP Cloud Identity Services acting as a proxy. The point, Ameling says, is to know “who did what, and the entire traceability of end-to-end business processes.”
The SAP AI Agent Hub, separately announced at Sapphire and now available to more SAP customers through Joule Studio, gives IT departments an inventory of every agent, LLM, and Model Context Protocol server running across the enterprise, SAP-built or otherwise. It also reports behavioral telemetry: goal completion rate, tool-call correctness, root-cause traces, and human-in-the-loop frequency.
The marketing language SAP is using around this is the “Autonomous Enterprise.” But to fulfill this vision, where humans are in the loop, but the gruntwork is done by agents, the agents need to be grounded in enterprise data and auditable.
The post SAP launches managed Joule Studio with Cursor and Claude Code support