Salesforce and other vendors are shifting from traditional per-seat SaaS pricing to complex models that charge based on AI agent work units, reflecting the growing automation of workflows by AI agents. This transition introduces pricing unpredictability and contractual challenges, making it essential for organizations to negotiate transparent, fair terms to manage costs and maximize AI benefits effectively.
Salesforce recently reported that its agent product reached an $800 million annual run rate, marking a 169% year-over-year increase, driven by billing based on “agentic work units” rather than traditional token counts. This shift reflects a broader industry trend where vendors like Microsoft and ServiceNow are moving away from simple per-seat pricing models to more complex meters that charge based on the actual work AI agents perform within their platforms. As AI agents increasingly automate workflows across multiple systems, companies must grapple with new pricing structures that go beyond model-building costs to include fees imposed by various platform providers acting as “toll booths.”
The traditional SaaS pricing model, which charged per user seat, is becoming obsolete as AI agents can perform tasks without occupying a human seat. Vendors are now introducing hybrid pricing models that combine seat licenses with additional charges for agent actions. Salesforce uses a credit system for agentic work units, Microsoft employs Copilot credits for different AI features, and ServiceNow charges based on operational work triggered by agents. These evolving pricing schemes introduce complexity and unpredictability, making it challenging for organizations to forecast costs and manage budgets effectively as AI usage scales rapidly.
Contractual and policy considerations are becoming critical in this new landscape. Vendors like SAP are imposing strict API usage policies that may restrict or heavily regulate how third-party AI agents interact with their systems, potentially limiting integration options or increasing costs. Organizations must proactively negotiate terms that clarify agent access, usage caps, billing units, and data handling to avoid unexpected expenses and ensure operational flexibility. Failure to address these issues early can lead to unfavorable terms, especially once AI agents become integral to business workflows and reduce human seat requirements.
A fair agent licensing model should be transparent, with clear usage meters, predictable pricing, and distinctions between different types of agent actions such as reading, writing, or executing tasks. Customers should have the ability to forecast usage, set spending caps, and export usage data. Conversely, rent-seeking models may obscure billing details, charge for failed actions, restrict third-party agents, and bundle credits with expiration, all of which can lead to inflated costs and vendor lock-in. Awareness of these dynamics is essential for buyers to negotiate effectively and avoid costly surprises.
Ultimately, the rise of AI agents is transforming the commercial unit of software from human seats to delegated agent work. Organizations must understand this shift to design cost-effective AI deployments and negotiate contracts that reflect the true value and scale of agentic workflows. Early and informed engagement with vendors about agent pricing, access, and impact on existing licenses is crucial to maintaining control over costs and maximizing the benefits of AI automation. Builders and decision-makers are encouraged to educate themselves on these new pricing models and contractual nuances to navigate the evolving AI software landscape successfully.