Enterprise software being repriced

Enterprise software is starting to look less like gym memberships and more like electric bills. For two decades, companies like Salesforce and Microsoft mostly charged by the seat, which meant one employee license usually equaled one monthly fee. Now both are pushing pricing that also counts what an artificial intelligence system actually does, from answering a customer question to completing a workflow. (salesforce.com) (microsoft.com) Salesforce has made that shift unusually explicit. Its Agentforce pricing page says customers can buy “consumption-based pricing” with Flex Credits at $500 per 100,000 credits, pay $2 per conversation for some customer-facing uses, or still buy user-based add-ons for employees. Salesforce also says Flex Credits are meant to align cost with “the business value your AI agents create,” and each action, like updating a record or summarizing a case, draws from that credit pool. (salesforce.com 1) (salesforce.com 2) Microsoft is moving in the same direction, but with a different mix. Microsoft 365 Copilot still has a per-user price of $30 a month, paid yearly, for employee use inside Microsoft 365. But Copilot Studio also uses a separate credit system for agents, with prepaid plans and pay-as-you-go billing for agents that run on websites, apps, or social platforms. Microsoft’s licensing documentation says the common unit for agents changed from messages to Copilot Credits on September 1, 2025. (microsoft.com) (learn.microsoft.com) That pricing change sounds technical, but it points to a bigger change in how software gets used. The old software model assumed a human employee opened the software interface, clicked buttons, and spent hours inside one vendor’s product. An artificial intelligence agent breaks that pattern because the worker may ask the agent for a result, and the agent may call several tools behind the scenes without the worker ever opening the original application. (fastcompany.com) (businessinsider.com) That is why the familiar metric of “daily active users” starts to lose some of its power. If a procurement team still pays Salesforce for customer data or Microsoft for workflow automation, but employees interact through an artificial intelligence assistant instead of the vendor’s own screen, the vendor can keep the revenue while losing direct attention. Fast Company framed that risk simply: customers may still buy from you, but they may no longer visit you. (fastcompany.com) That changes what buyers have to measure before signing a contract. A seat-based purchase is easy to model because a chief information officer can count 5,000 employees and multiply by a license price. A credit-based or outcome-based purchase forces the buyer to estimate how many actions an agent will take, how often it will fail, which systems it must connect to, and whether the savings are bigger than the bill. Microsoft now offers an agent usage estimator for Copilot Studio, which shows how central forecasting has become to the sale itself. (microsoft.github.io) (learn.microsoft.com) It also changes what makes one software platform defensible. In the seat era, a sticky user interface and broad employee adoption were often enough to protect a vendor. In the agent era, the harder moat may be the application programming interface, which is the software doorway another program uses to get in, plus the data permissions, audit logs, uptime guarantees, and policy controls that let a company trust an agent in production. (fastcompany.com) (learn.microsoft.com) Consultants and analysts are already describing this as a repricing, not a minor add-on. Bain said hybrid pricing models that mix per-seat charges with artificial-intelligence-based usage or outcome metrics have become the dominant interim strategy. Bain also said vendors now have to align pricing with perceived customer value instead of relying on the old one-size-fits-most seat model. (bain.com) There is a reason vendors are not abandoning seats all at once. Per-seat pricing is predictable for finance teams, and pure usage pricing can scare buyers if monthly costs swing with every burst of agent activity. That is why both Salesforce and Microsoft currently offer mixed menus: a base subscription for people, then metered charges for agent work that expands outside the old boundaries of a human user license. (salesforce.com) (microsoft.com) (bain.com) The stock market has noticed the uncertainty before many customers have. Business Insider reported on April 8, 2026, that software executives argued artificial intelligence would not eliminate the need for platforms like Salesforce and Microsoft, even as investors questioned what happens if software becomes less visible to end users. The anxiety is not just about whether companies still need software; it is about whether the old pricing logic survives when software starts acting more like labor than like a digital filing cabinet. (businessinsider.com) The practical result for buyers is a new checklist. A company comparing enterprise software in 2026 has to ask how an agent connects to the system, what one completed task costs, how usage is capped, what happens when the model makes a mistake, and whether the vendor can prove a business outcome like faster case resolution or higher sales conversion. Those questions sit closer to cloud infrastructure buying than to the old software question of how many employees need logins. (salesforce.com) (learn.microsoft.com) (bain.com) The deeper shift is that software is being judged less by presence and more by output. If an artificial intelligence agent can finish a refund, qualify a sales lead, or summarize a legal file without a worker opening the vendor’s interface, the winning product may be the one with the best plumbing rather than the prettiest screen. In that world, pricing follows the same logic: not how many people touched the tool, but how much useful work the