AINEWS
Salesforce is betting the future of enterprise software isn't seats or credits but outcomes. The harder problem is that 'resolution' is a vendor defined state.
Salesforce's newest product costs $2 every time its AI agent closes a support ticket without human help. It costs nothing when the same agent fails, gets a thumbs-down, or hands the conversation to a person. That is the entire business model of Agentforce Help Agent, which went generally available in July 2026 as the company's third pricing experiment in roughly eighteen months and the most aggressive bet yet that enterprise software should be billed by the work it does, not the humans who use it.
The bet has structural reasons behind it. A seat license assumes a person at a keyboard, generating value roughly in proportion to the hours they log in. An AI agent that resolves a ticket works differently. It might run for thirty seconds or thirty minutes, charge a fraction of a cent in tokens or several dollars in cloud compute, and the customer can only know it worked because the ticket closed. Salesforce's pitch is that the same telemetry on its own help.salesforce.com site, where Agentforce has handled 4.3 million inquiries and resolved 70 percent of them, is the implicit ROI case for the $2.
That number is a Salesforce claim, not an independent benchmark. It is also the lever the entire model pivots on, because "resolved" is a Salesforce-defined state. The agent decides when the conversation qualifies, the customer has limited visibility into how the call was made, and analysts have already flagged the resulting margin of error. Constellation Research described pay-per-resolution as "based on lessons from customer implementations" after a per-conversation model proved too crude, then added the warning every enterprise buyer should hear out loud: "there will likely be haggling over definitions and fuzzy math."
Help Agent is not Salesforce's first try. At Dreamforce 2024 the company sold agents at $2 per conversation. In May 2025 it pivoted to Flex Credits, a metered bundle where 20 credits cost a cent per text action and 30 credits cost 1.5 cents per voice action, sold in 100,000-credit packs for $500 alongside a Flex Agreement that converted existing licenses. The architecture now stacks three layers: seats for the humans, Flex Credits for the agent actions, and per-outcome fees for Help Agent. The Register framed the result as "an anxiety-filled architecture" of "Seats, Flex Credits, and all-you-can-eat contracts." Salesforce has not publicly disputed the characterization.
Practitioners are already running the gaming math. Salesforce Ben flagged a path where a user claims dissatisfaction with a closed ticket, the agent reopens it, and Salesforce eats the token cost without billing a resolution. Ian Gotts of Elements.cloud, quoted by Salesforce Ben during the Flex Credits rollout, said customers "don't want to hand out a blank cheque." Steve Snell of Attentis compared the credit bundle to "a 1991 cell phone data plan." Those reactions land because the meter is opaque. Data 360 and Agentforce processing are unmetered during a Help Agent interaction; $2 is the only billable layer, which means the customer is paying for a Salesforce-defined state on a Salesforce-controlled ledger.
The broader industry is heading the same direction. Oracle, HubSpot (with Breeze), and Pegasystems have all pitched outcome-based AI pricing in 2025 and 2026, framing it as the answer to runaway inference cost. Salesforce's parallel metric, Agentic Work Units (AWUs), reached 3.8 billion lifetime, with 1.6 billion in Q1 alone (up 111 percent quarter over quarter) and Service Agents at 232 million (up 66 percent QoQ). AWUs are not billed, but they let Salesforce talk about scale without committing to a single price. The Help Agent launch narrows the question: is Salesforce willing to bill only on the work that actually produced an outcome, and is the customer willing to trust Salesforce's definition of "outcome" enough to write that check?
The first year of pay-per-resolution will be the test. If the resolution audit trail holds up under customer scrutiny, Flex Credits could migrate into a per-outcome default across more of the Agentforce surface, and the seat might finally start to shrink. If haggling over fuzzy math becomes a contract-line item, expect a slower transition: seat licenses persisting for years, outcome fees confined to narrow use cases, and Salesforce left defending the meter rather than the product.