Few pricing stories in enterprise software this year illustrate the industry’s uncertainty about how to charge for AI agents better than Salesforce’s Agentforce. In roughly eighteen months, Salesforce has shipped three fundamentally different pricing models for the same product line — and the latest, generally available since July 2026, flips the entire billing logic on its head by charging customers only when an AI agent actually resolves a problem.
From per-conversation to per-action to per-outcome
When Salesforce first unveiled Agentforce at Dreamforce in the fall of 2024 — rebranding what had previously been the Einstein Copilot project — it chose a simple usage-based metric: $2 per conversation. The pitch was straightforward, much like paying per chat session or per case. But the model had an obvious flaw that customers surfaced quickly: it billed the same $2 whether the AI agent successfully resolved a customer’s issue or the customer got frustrated and demanded a human. Complaints centered on two recurring themes — being billed regardless of outcome, and being billed again for escalations that failed entirely.
Salesforce’s response, introduced in May 2025, was Flex Credits: a more granular consumption model priced at roughly $0.10 per discrete action rather than per full conversation. That addressed some of the coarseness of the original metric but introduced its own complexity, since the total cost of any given workflow now depended on how many individual actions — lookups, tool calls, generations — a task required, rather than a flat per-interaction rate.
Now, as of July 2026, Salesforce has layered in a third structure: per-user licensing starting at roughly $125 per user per month, which CEO Marc Benioff has framed as pricing for “digital labor” — treating AI agents more like additional headcount than like software features. Running in parallel with this, Salesforce’s newest release, Agentforce Help Agent, became generally available in July 2026 with pay-per-resolution pricing: a business is billed only when an issue is fully solved end-to-end without human intervention, not per conversation and not per action.
Three pricing models, one product line
What makes this notable is that all three models are live simultaneously, applied to different parts of the same Agentforce family depending on use case and buying structure. The $2-per-conversation rate still applies specifically to fixed-model customer-facing agent deployments. Flex Credits, sold in blocks such as $500 for 100,000 credits, cover customer, employee, and voice use cases under the consumption model — though Salesforce has structured its systems so that Flex Credits and per-conversation billing cannot run in the same org simultaneously, forcing customers to commit to one metering approach per deployment. On top of both, Agentforce 1 Edition bundles unmetered usage within defined scope at roughly $550 per user per month, aimed at organizations that want predictable costs rather than exposure to consumption swings.
Industry commentary on the shift has been split between reading it as strategic experimentation and reading it as pricing chaos. One widely circulated framing in the SaaS commentary community put it bluntly: on paper, three pricing models for one product in eighteen months looks like an inability to settle on a strategy. But viewed differently, it may be Salesforce hedging across every plausible way an AI-agent market could ultimately price itself, rather than betting the entire product line on one model too early.
The real cost of getting started is higher than the headline number
The $2-per-conversation figure that still anchors much of Salesforce’s public messaging turns out to describe only a narrow slice of what most enterprise buyers actually pay. Agentforce requires a mandatory Data Cloud subscription, typically running north of $108,000 per year on its own, before the agent platform can meaningfully function, plus implementation services and knowledge-base setup work. Consulting firms that have implemented Agentforce for clients across service, sales, and healthcare estimate genuine first-year costs for a mid-market company landing anywhere between $150,000 and $600,000, once Data Cloud, Flex Credits, per-conversation billing, and per-user licensing are all accounted for together — a considerably more complex picture than the entry-level sticker price suggests.
Outcome-based pricing is spreading beyond Salesforce
Salesforce is not alone in wrestling with how to price autonomous AI agents, and it may not even be the most aggressive mover. Zendesk has adopted its own outcome-based structure for AI agents, charging roughly $1.50 per automated resolution on committed plans, or $2 on pay-as-you-go terms — pricing philosophically aligned with, though structured slightly differently from, Salesforce’s new Help Agent model.
The starkest example of pure outcome-based pricing, however, comes from Sierra AI, the customer-service agent startup co-founded by former Salesforce co-CEO Bret Taylor. Sierra charges customers only when its AI agent successfully resolves an issue without human involvement — nothing for attempts that fail or escalate. Taylor has described the model as treating AI agents the way a business treats a commissioned salesperson: compensation tied directly to results rather than activity. The approach has scaled quickly: Sierra reportedly reached $100 million in annualized recurring revenue within 21 months of launch, followed by a $50 million quarter that pushed it past $150 million in ARR by early 2026, at a reported $10 billion valuation. Notably, roughly one in four Sierra customers reportedly has annual revenue exceeding $10 billion, and some customers report between 50 and 90 percent of service interactions being fully automated without human handoff.
List price increases compound on top of the new consumption layer
Agentforce’s pricing evolution is unfolding against a backdrop of separate, more conventional list price increases across Salesforce’s core CRM products. Salesforce held prices flat from 2016 through 2023, then raised list prices twice: 9 percent in 2023, followed by 6 percent in 2025, both applied to Enterprise and Unlimited editions of Sales Cloud, Service Cloud, Field Service, and select Industry Clouds. Layered independently on top of list pricing, most Salesforce order forms include a contractual annual uplift clause compounding at 8 to 10 percent per year — a mechanism distinct from list price changes that applies directly to a customer’s already-negotiated contract rate at each renewal.
The combined effect creates what licensing advisors are calling a “double escalation” problem for customers renewing in 2026: those on contracts signed in 2022 or 2023 face compounded annual uplift applied every year since signing, plus a new, higher list price that resets the reference point for any renegotiated discount — meaning a 1,000-user Enterprise deployment signed in 2020 could see its annual contract value climb 15 to 40 percent by 2026, depending entirely on how the uplift clause was negotiated at each renewal along the way.
What this means for buyers evaluating agentic AI
Procurement advisors reviewing Salesforce renewals are urging customers to stop evaluating discounts as a percentage off list price and instead compare the actual proposed dollar renewal rate against their previous contracted rate plus a defensible increase — treating the new, higher list price as a negotiating anchor rather than an accepted baseline. For any organization evaluating Agentforce specifically, the added advice is to model total first-year cost across Data Cloud, the chosen usage-billing mechanism, and implementation, rather than anchoring expectations to the $2-per-conversation figure that dominates public marketing.
More broadly, Salesforce’s willingness to run three pricing models concurrently — and the parallel moves by Zendesk and Sierra toward paying strictly for resolved outcomes — suggests the enterprise software industry has not yet converged on a single accepted way to price autonomous AI agents. For CIOs and procurement teams, that means agentic AI contracts in 2026 and 2027 are likely to require more active management, more scrutiny of consumption mechanics, and less reliance on any single headline price than traditional per-seat software licensing ever demanded.
Why outcome-based pricing appeals to buyers — and why it’s hard to implement
From a buyer’s perspective, the appeal of outcome-based pricing is obvious: it aligns the vendor’s incentives directly with the customer’s desired result, rather than with raw usage volume. A support organization paying per resolved ticket has no reason to worry that an AI vendor is financially incentivized to keep conversations running longer or to encourage unnecessary escalations, since revenue only flows on genuine success. That alignment is a meaningful improvement over the earlier per-conversation model, where Salesforce was, in effect, being paid the same whether or not its own product worked.
The practical difficulty is defining “resolved” in a way both vendor and customer trust. A ticket that a customer abandons without complaint, a case that gets resolved days later through an unrelated channel, or an issue that the AI agent believes it solved but the customer disputes, all create measurement disputes that per-conversation or per-action billing never had to confront. Vendors moving toward outcome-based models are, in effect, taking on new operational responsibility for building auditable, mutually agreed definitions of success — a burden that per-seat and per-conversation pricing never required them to carry, and one that is likely to generate its own category of contract disputes as outcome-based agreements scale.
What to watch next
Given how quickly Salesforce has already cycled through pricing models, procurement teams evaluating Agentforce should treat the current structure as a snapshot rather than a stable long-term arrangement. Contract language that locks in favorable per-resolution or Flex Credit rates for a defined term, rather than assuming today’s published pricing will still apply at the next renewal, is likely to matter more for Agentforce customers over the next two years than it has for any previous generation of Salesforce licensing.