SAP consultants are raising questions about whether a recent Salesforce demonstration of its AI technology at the Dreamforce conference last week would breach the German software giant's API access policy if deployed in real-world production systems. The demo, presented during a keynote in San Francisco, showed an AI agent onboarding suppliers within SAP's system while users worked in Slack. Salesforce has denied the demonstration would break any of SAP's access rules, but independent experts contend the approach conflicts with restrictions SAP put in place five months ago.

During the presentation, Salesforce product marketing director Katie O'Neil showcased an AI agent named Marshall working inside an SAP sandbox environment, where it learned "where to click, which fields are required" and used reasoning to understand business rules behind supplier onboarding processes. The agent would package these actions into a library of trusted steps, allowing users in Salesforce's Slack platform to onboard customers in SAP through what O'Neil described as "deterministic execution" rather than improvisation. The demo featured global engineering manufacturer Siemens as a customer example, though the partnership between Salesforce and SAP was not the result of any new agreement between the two vendors.

Mario de Felipe, VP of SAP data and AI at IT services company Apiphani, argued in a social media post that the demonstration would not be permitted under SAP's API Policy if replicated in production environments. He pointed out that the policy forbids "impersonation techniques" and requires any autonomous AI to go through SAP "endorsed architectures" to access SAP applications. According to de Felipe, Salesforce didn't simply connect to SAP's API to update a supplier but instead "connected to an SAP system, captured SAP application logic… how the BP [business process] role model works... the complete SAP's product behavior." A Salesforce spokesperson responded by noting that the demonstration "interacted with SAP through its user interface using service accounts, without using SAP APIs," meaning the API restrictions cited don't apply to the activity shown. SAP declined to comment on the matter.

The controversy stems from SAP's API Policy launched in April, which prohibits API use for "interaction or integration with (semi-) autonomous or generative AI systems that plan, select, or execute sequences of API calls" unless done "through and within the limits of SAP-endorsed architectures." Critics have characterized the policy as protectionism designed to funnel customers toward SAP's own AI tools, particularly its Joule digital assistant and the Joule Agent Builder released in December 2025. One senior analyst speaking off the record said SAP's first AI effort is "only a digital assistant" and that linking AI access to RISE with SAP—the vendor's cloud migration partnership—was "a big mistake" that denied access to roughly 93 percent of large customers. Independent consultant Marian Zeis noted the policy is "reasonable as a governance document, but it still does not fit very well with how customers actually work today," while former SAP partner CEO John Appleby suggested demos are often "designed to be edgy but not break any license agreements" and that Salesforce's use of Siemens' name suggests compliance was verified.

The dispute highlights a broader battle over which company controls the user experience layer as enterprise software moves toward AI-driven workflows. SAP would prefer customers build agents within its own environment that then reach out to other systems, a strategy it emphasized at its Sapphire conference in May when it said Joule Studio 2.0 would let developers connect to non-SAP applications and third-party agents. Whether publishing an API policy document will be sufficient to prevent competitors like Salesforce, ServiceNow, or Microsoft from capturing the end-user relationship remains an open question, particularly as most SAP customers who have already deployed generative AI using APIs may now be non-compliant according to the new policy. The practical challenge for SAP is that enterprises won't necessarily wait for endorsed architectures when they can integrate AI capabilities through the platforms where their employees already spend their workdays.