Noise 3 min read

SAP Locks Down Its Agent Ecosystem. Salesforce Opens Its Up

SAP Locks Down Its Agent Ecosystem. Salesforce Opens Its Up

Why We're Watching This: SAP and Salesforce are taking opposite approaches to enterprise agent access. Which model wins will determine how much control enterprises retain over their AI stack for the next decade.

Key Takeaways

  • SAP is acquiring Prior Labs for structured data AI and investing €1 billion over four years, while blocking all agents except those using SAP-endorsed architectures including Nvidia’s NemoClaw
  • Salesforce is taking the opposite approach with Headless 360, exposing its entire platform as APIs, MCP tools, and CLI commands accessible to any agent including OpenClaw
  • Nvidia’s NemoClaw sits in the middle as the common enterprise agent infrastructure layer endorsed by both SAP and Salesforce
  • Prior Labs’ TabPFN model series has been downloaded over 3 million times and tops the TabArena benchmark for tabular AI
  • As of February 2026, OpenAI’s own COO said enterprise AI has “not yet really penetrated business processes” at scale

SAP and Salesforce are both making major enterprise AI moves this week, and they are pointed in opposite directions. The divergence matters more than either announcement individually.

SAP announced it has entered into a definitive agreement to acquire Prior Labs, an 18-month-old German startup building tabular foundation models (TFMs) purpose-built for structured data. SAP will invest €1 billion over four years to scale it into an independent frontier AI lab. Prior Labs’ TabPFN model tops the TabArena benchmark and has been downloaded over 3 million times.

While building its own AI lab, SAP has simultaneously locked down its agent ecosystem. The company’s API policy prohibits AI agents from accessing its products except through SAP-endorsed architectures, which currently means SAP’s own Joule Agents and Nvidia’s NemoClaw, the enterprise-grade version of OpenClaw built on Nvidia’s Agent Toolkit. Salesforce and its Agentforce are blocked. OpenClaw is blocked unless accessed via NemoClaw.

Salesforce is doing the opposite. Its new Headless 360 architecture exposes the entire Salesforce platform as APIs, MCP tools, and CLI commands that any agent can access, including OpenClaw, Claude Code, Cursor, and Codex. Salesforce co-founder Parker Harris framed it directly: “Why should you ever log into Salesforce again?” The platform is being rebuilt for agents, not human browsers, as enterprise agent deployment timelines compress across the industry.

We have not yet really seen AI penetrate enterprise business processes. — Brad Lightcap, COO, OpenAI

Nvidia’s NemoClaw is the infrastructure play that sits beneath both. Built on Nvidia’s Agent Toolkit, it provides the open source runtime, policy-based security guardrails, and enterprise deployment framework that both SAP and Salesforce are building around, part of a broader shift toward agents replacing apps as the primary interface layer. Adobe, Atlassian, Cisco, CrowdStrike, ServiceNow, and Siemens are all working with Nvidia’s Agent Toolkit. Nvidia is positioning itself as the picks-and-shovels layer regardless of which platform strategy wins.

SAP’s closed-agent stance creates a direct procurement constraint. Enterprise teams running workflows across both SAP and non-SAP systems need to validate that their chosen agent framework is on SAP’s endorsed list before building any cross-platform automation. OpenClaw is not authorised in SAP’s ecosystem unless accessed via NemoClaw.

SAP’s stock dropped as much as 16% in January after cloud backlog growth disappointed, reflecting the pressure the “SaaSpocalypse” is placing on legacy enterprise software vendors. The Prior Labs acquisition and the agent lockdown are both defensive moves in that context: build a proprietary AI capability edge while preventing competitors from accessing customer data through third-party agents.

The SAP vs Salesforce split is the early shape of a broader enterprise AI infrastructure battle. Closed and controlled on one side. Open and interoperable on the other. Which architecture enterprises standardise on will determine how much flexibility their AI teams have for the next several years, a shift tracked on Relve, an AI trends intelligence platform.

Neelam Khan

Neelam Khan

Verified

Lead Editor

Neelam Khan is a Lead Editor at Relve, covering AI news, tools, product updates, search trends, and business use cases. She filters noise from useful signals for founders and teams, drawing on her previous work in AI SEO, content strategy, and tool research with Wellows and AllAboutAI.

Read Full Bio →