The pricing model that assumed a human was clicking
Per-seat SaaS pricing has held for roughly two decades on a simple assumption: a licensed human user logs in, clicks through screens, and generates a predictable, bounded volume of system activity that scales roughly with headcount. AI agents break that assumption cleanly. A single AI agent can trigger thousands of API calls in a day without ever occupying a licensed seat, meaning the traditional metric platforms used to size and price access no longer bears any reliable relationship to actual system usage or load.
That mismatch is forcing every major enterprise platform vendor to make a real strategic choice, not a minor pricing tweak: either meter and charge for the new usage pattern directly, or restrict it to protect the existing seat-based revenue model. ServiceNow, Workday and SAP have each picked a different answer, and the divergence itself is revealing about how each company weighs new revenue opportunity against the risk of alienating customers and integration partners.
ServiceNow's tollgate: pay per action, not per seat
ServiceNow COO Amit Zavery announced usage-based metering at the company's Knowledge 2026 conference, introduced through a new integration layer called Action Fabric. Under this model, customers pay based on how many operations an external AI agent completes through that layer, with Anthropic's Claude serving as the launch partner given direct integration capability into the platform.
ServiceNow EVP Jon Sigler articulated the broader ambition behind the mechanism: 'we're going to have this universal action layer, where all of these systems are calling directly into our Action Fabric.' That framing positions Action Fabric as infrastructure ServiceNow wants to become the mandatory routing layer for any AI agent, regardless of vendor, that needs to touch data stored inside ServiceNow's applications, well beyond a simple pricing mechanism.
The tax framing that is hard to shake
JPMorgan analyst Mark Murphy's characterization, that this is 'effectively a tax on customers using outside AI agents to interact with data they already store in ServiceNow's apps,' captures the core tension precisely. Customers already pay ServiceNow for the platform and already own the data inside it; the new charge applies specifically to the act of an external tool reaching in to use that data through an agent rather than through a human clicking the ServiceNow interface directly.
That framing matters for procurement negotiations going forward. Enterprise buyers evaluating ServiceNow contracts should expect to model agent-driven API volume as a new, variable cost line item alongside traditional per-seat licensing, and should push specifically for volume-based pricing tiers or caps during renewal negotiations rather than accepting open-ended per-action billing that could scale unpredictably as agentic workflows expand across the organization.
Workday follows the same playbook
Workday CEO Aneel Bhusri identified 'considerable financial upside' in adopting a broadly similar metering approach to ServiceNow's, describing it explicitly as a strategy to capture new revenue from agent-driven access rather than treating it as a cost center to absorb. That two of the three major platforms independently arrived at metered, tollgate-style pricing for agent access suggests this is becoming the default industry pattern rather than an isolated experiment from a single vendor.
For customers running workloads across both ServiceNow and Workday, the practical effect compounds: agent-driven automation that spans both platforms, a common pattern for HR and IT service workflows that touch both systems, will now carry metered costs at each hop rather than a single predictable subscription fee covering the full workflow.
SAP chose restriction over metering, and paid for it in goodwill
SAP took the structurally different path. An April 2026 policy update prohibits third-party AI agents from planning, selecting or executing sequences of API calls without SAP's explicit authorization, with SAP's own Joule Agents exempted and permitted unrestricted access. CEO Christian Klein defended the policy on customer-ownership grounds, stating customers 'would not pay to access their own data,' framing the restriction as protecting customers from exactly the kind of metered tax ServiceNow and Workday are now charging.
But the restriction drew immediate pushback from SAP's own user group, DSAG, along with partners whose products rely on connectors to tools like Microsoft Copilot and Salesforce Einstein, since those integrations now require SAP authorization that did not previously exist as a gating step. The policy protects customers from a metering tax while simultaneously constraining which third-party AI tools they can freely use inside SAP environments, trading one friction point for another rather than eliminating it.
What finance teams need to build now
The shift from predictable per-seat subscription costs to variable, consumption-based agent billing is a structural change to enterprise software budgeting, not a minor line-item adjustment. Finance teams accustomed to forecasting SaaS costs based on headcount now need visibility into projected agent activity volume, a metric most organizations have never had to track or forecast before this shift began.
The practical response is building consumption monitoring and alerting into agentic AI deployments from the start, treating API call volume the way cloud teams already treat compute and storage consumption: as a metered resource requiring active cost governance, budget alerts and usage caps, rather than assuming agent-driven automation carries the same flat, predictable cost profile as the human-driven workflows it is replacing.
The precedent this sets for every other platform vendor
Datadog's own response, capping MCP server usage at 5,000 daily or 50,000 monthly requests with exceptions available, represents a third distinct model: consumption caps rather than either open metering or outright restriction. That three major platform categories, ITSM, HR and infrastructure monitoring, have each landed on a different answer to the same underlying problem suggests the industry has not yet converged on a standard, and enterprise buyers should expect continued experimentation and renegotiation of agent access terms across their broader vendor portfolio over the next several contract cycles.
CIOs building 2027 software budgets should treat agent-access pricing as an open, unresolved variable rather than a settled cost, and should use contract renewal cycles specifically to negotiate explicit terms now, before vendor pricing models harden further and negotiating leverage shifts more decisively toward the platform vendors that control the data agents need to reach.
