From User Seats to Agent Operations: A New Licensing Baseline
Enterprise AI agents are forcing a fundamental rethink of licensing. Traditional models assume human users who log in, complete tasks and log out. Agentic AI, by contrast, can fire off thousands of API calls in a single workflow, touching data, triggering automations and updating records without ever appearing as a named user or seat. That gap is now closing as major vendors reframe AI agent licensing around consumption rather than headcount. For IT leaders, this shift has immediate implications: what looked like low-friction, low-cost automation can quickly translate into a metered service with its own budget line and governance model. As enterprise AI agents move from experiment to core infrastructure, software vendor pricing strategies are increasingly built around controlling, measuring and monetising this high-volume, machine-driven access to their platforms.
ServiceNow’s Action Fabric: Metering External Enterprise AI Agents
ServiceNow is taking the most explicit step toward consumption-based AI agent licensing. The company has introduced Action Fabric, a mandatory intermediary layer that all external AI agents must traverse to access ServiceNow data and workflows. Instead of tying access to traditional user licenses, ServiceNow will bill per operation an agent completes, effectively turning agentic AI access into a metered utility. For enterprises that already wired third-party tools like Anthropic’s Claude into ServiceNow, this means existing automations now carry a variable cost that did not exist when they were designed. Technically, Action Fabric gives ServiceNow better observability and control over agent behaviour. Commercially, it establishes a template for software vendor pricing that treats AI agents as high-frequency platform consumers whose access must be explicitly licensed, governed and budgeted.
SAP Tightens Architectures: Policy, Control and Vendor Lock-In
SAP is reshaping AI agent access through policy rather than a new product. An updated API policy restricts third-party AI agents from autonomously chaining API calls inside SAP systems unless they operate via SAP-approved architectures. SAP’s own Joule Agents are explicitly allowed, but connectors feeding Microsoft Copilot or Salesforce Einstein do not currently qualify. That places many existing integrations in a grey zone and potentially out of compliance with SAP’s terms. While SAP’s leadership has publicly promised not to charge customers for access to their own data and to keep the architecture open, the written policy language is more restrictive. The net effect is strategic: SAP is asserting tighter control over how agentic AI access is designed and who brokers it, nudging customers toward its native agents and partner models, and away from ungoverned third-party automation.
Workday Signals Monetisation: Preparing for New Pricing Playbooks
Workday has not yet rolled out a concrete product or tariff for AI agent licensing, but its leadership has clearly flagged monetising agent access as a significant future revenue lever. In HR and finance environments, where routine, rules-based processes are ripe for automation, enterprise AI agents are proliferating quickly. That growth makes Workday’s stance particularly consequential: any eventual move toward metered operations, dedicated AI agent SKUs or tiered access could materially change total cost of ownership for large deployments. Even without formal changes in place, Workday’s signalling gives IT and procurement teams advance warning. Organisations should anticipate that AI agent access to core systems will not remain a free by-product of existing licenses, and that software vendor pricing will increasingly distinguish between human users and automated, high-volume agent workloads.
Strategic Responses: Rethinking Vendor, Licensing and Automation Roadmaps
These shifts collectively mark a consolidation of power around agentic AI access. ServiceNow, SAP and Workday are making it clear that enterprise AI agents touching their platforms will be governed, metered or architecturally constrained on their terms. For organisations, this is both a risk and a forcing function. The first step is an audit: catalogue which enterprise AI agents are calling which systems, through which APIs, under what contracts. Many integrations built before these policy changes may now be non-compliant or unexpectedly costly. The second step is to embed AI agent licensing into vendor strategy. New contracts and renewals should explicitly address agent operations, telemetry, throttling and pricing. Finally, automation roadmaps must assume platform-dependent economics: some processes may be cheaper or more sustainable if redesigned around vendors’ native agents rather than external tools.