ServiceNow, SAP, and Workday Are Making Three Different Bets on Whether to Charge AI Agents
At its Knowledge 2026 conference in Las Vegas, ServiceNow launched Action Fabric, a metered integration layer that charges external AI agents by the operation every time they interact with ServiceNow data. Claude is a launch partner. The pitch is straightforward: if an AI agent is going to run thousands of automated actions against your platform, that usage should be billed the same way API calls or seat licenses are billed.
Workday hasn't built the equivalent yet, but its CEO has said publicly that charging AI agents for access offers considerable financial upside for the company. That's a company signaling where it expects to land, not one describing something already shipped.
SAP went the other direction entirely
SAP updated its API policy in April 2026 to actively block third-party autonomous agents from reaching its systems outside of SAP's own endorsed architectures. CEO Christian Klein's public argument is that customers shouldn't have to pay extra just to access their own data through an agent, framing metered agent access as a tax on customers rather than a legitimate new product line.
Three bets on what an AI agent actually is
Line these three up and they aren't small variations on the same strategy, they're genuinely different theories of what a third-party AI agent represents to an enterprise SaaS incumbent. ServiceNow is treating agent access as a new revenue line to build and meter today. Workday is treating it as a revenue line worth building toward once the market matures. SAP is treating it as a threat to platform control, something to wall off rather than monetize, at least on anyone's terms but its own.
Why this matters if you're buying any of these platforms
If your organization is planning to connect Claude, or any other AI agent, to ServiceNow, SAP, or Workday, the economics of that connection now depend entirely on which vendor you're dealing with. A metered-access model like Action Fabric means your AI agent's usage becomes a line item that scales with how much automation you actually run, which is a real cost to model before committing to an agent-heavy workflow. SAP's blocking policy means the agent architecture has to be SAP's own, not whatever third-party tool your team already prefers. None of these three approaches is obviously wrong. They're just incompatible with each other, and worth checking before you build a workflow that assumes one of them.