SAP's AI strategy: Come for the openness, stay because you have to

Like many enterprise software vendors, SAP's share price plunged during the "SaaSpocalypse" - the belief that GenAI and vibe-coding could disrupt traditional enterprise app vendors. At its annual conference in Orlando last week, the ERP giant pushed back with a new vision for how GenAI will work across its enterprise apps and analytics portfolio. On the one hand, it is helping users build agents based on data from outside the SAP ecosystem. On the other, it is arguably creating friction for those wanting to build agents on third-party platforms and use data from SAP systems to power them. At its Sapphire conference, SAP announced Joule Studio 2.0, with new features allowing developers to create and manage AI agents. Agents created in Joule Studio will natively support Model Context Protocol and A2A protocols - two standards designed to help GenAI integration between data sources - crucially allowing the SAP tool to connect and collaborate with third-party tools and agents. Other features, such as the agentic orchestration, are also designed to run across hybrid landscapes, while real-time data ingestion promises to support "context-aware processes" across SAP and third-party systems. Speaking to the conference, Muhammad Alam, SAP executive board member for product and engineering, said: "Underpinning the autonomous suite are out-of-the-box agents, hundreds of agents cutting across all core business processes. These agents come together into what we call assistance, or Joule assistants. "We've made extensibility a core design principle. You can extend any of these agents by adding tools, workflow steps, and even code through the same simple experience in Joule Studio. This allows you to connect them to your non-SAP applications, because we know you're going to have to do that." But this approach to agents roaming freely across the application estate might not support agents built and managed on other platforms. Control and access: The API policy critics Gartner senior director analyst Christian Hestermann said SAP's API policy, published last month, can be read as an effort to control access to capabilities inside the SAP platform and control how third-party AI platforms might build agents based on SAP's applications. "It's about more than just data; agents are expected to do not only read or manipulate data but do entire and maybe complex business activities or even chains of activities. SAP is trying to channel how and who can access the SAP systems through third-party AI platforms and solutions. If they do endorse these environments, SAP will charge customers extra for that," he said. As well as the agent integration in Joule Studio 2.0, SAP has announced a partnership with Anthropic to bring the Claud model into its SAP Business AI Platform. Hestermann said: "We think that the two things - the API policy and the Anthropic partnership - must be seen jointly. SAP is closing the door to third-party AI environments, especially agentic environments, but at the same time, sort of offering Anthropic Claude as something within a 'walled garden' of SAP." The game plan of positioning SAP at center of agentic AI development can also be seen in the data strategy, which the vendor reheated with the launch of Business Data Cloud (BDC), a partnership with Databricks from last year, which promised bi-directional data sharing between SAP Business Data Cloud and third-party data platforms. SAP has sought to strengthen its technology for data sharing across common enterprise sources outside its portfolio with the acquisition of lakehouse vendor Dremio and metadata company Reltio. "They started opening the landscape in Business Data Cloud last year, first with the Databrick partnership, and acquisitions of Reltio and Dremio, those two things go hand in hand. You can, through certain technologies, reach out to other landscapes, but it's more difficult to put a neutral AI layer on top of your entire [SAP] landscape," Hestermann said. Not the only game in town SAP faces competition, though. A bunch of other enterprise application vendors want their users to see them as the locus of control for agentic AI working across their mix of applications from various providers. Salesforce and its office collaboration platform Slack and Oracle both pitch their technology in this way, and so does enterprise workflow vendor ServiceNow. Whether these approaches work or not on a technical basis is the wrong question, said Faram Medhora, principal analyst for Technology Architecture and Delivery at Forrester. "The technical question is solved. The economic question is not. SAP's Joule Studio can reach into Salesforce through open protocols. Salesforce's Agentforce can call SAP. The protocols work. What remains unsolved is who pays for the runtime, who governs the agent, who owns the audit trail, and whose roadmap dictates what the cross-vendor agent can do next quarter." Large enterprise IT departments may be under pressure to execute an agentic AI strategy, but Medhora warns against viewing these choices only from a technical perspective. "The cross-vendor agent question is not a technology choice. It is a 2028 contract negotiation being decided in 2026, under the cover of a 2027 implementation timeline. The platform an enterprise picks to build its first cross-vendor agent becomes the platform that prices its entire AI estate for the next decade," he said. SAP wants to be that choice among its customers. It may make sense if the vast majority of their enterprise applications come from the German vendor, but the reality is, most large enterprises rely on a mix of application vendors, some of whom also want to be the main center for AI agents. While the technology may allow AI agents to work across these barriers, the commercial reality presents another challenge altogether. (R)