Salesforce’s AI Agents Turn Per-Seat Licensing Into a Moving Target

Salesforce’s deputy CFO has a name for the way his company currently charges for AI: an “anxiety-filled architecture.” Mike Spencer, who serves as deputy CFO and head of finance, used that phrase at the Deutsche Bank Technology Conference in August to describe Salesforce’s simultaneous experiments with three different ways to bill customers for AI agents. Weeks later, at Salesforce’s Dreamforce conference, the company unveiled AIforce, a new interface layer meant to let AI agents do Salesforce work through Slack, Claude, and other channels instead of a browser. The launch reopened the question Spencer had already admitted the company hasn’t answered: how do you charge for software when a human is no longer the one using it?

Why Per-Seat Licensing Breaks Down When Agents Do the Work

Salesforce built its business on a simple unit: the named seat. A company paid for each employee who logged in and used the CRM. That model assumes a human occupies the seat. AIforce, powered by what Salesforce calls its Headless Toolkit, makes the company’s data, workflows, and business logic available through any AI interface, not a fixed screen a person clicks through. When an AI agent or an API calls that logic directly, there’s no named user to license.

Bill Patterson, Salesforce’s executive vice president and general manager of CRM applications, described the shift plainly at an investor webinar earlier this month. Salesforce is devising, in his words, “a new pricing structure that really aligns to the benefits that customers realize from this new technology,” as The Register reported. That is a polite way of saying the old unit of measurement, the seat, no longer maps cleanly onto how the product gets used.

Eighteen Months of Pricing Experiments

From $2 a Conversation to Pay Per Resolution

Salesforce first put a price on its AI agents at Dreamforce 2024, with a flat rate of $2 per conversation. Customers and partners quickly complained that the model was hard to predict: a conversation could mean a fully resolved customer issue or a single unhelpful exchange, and the bill looked the same either way. By April 2025, Ian Gotts, CEO of the Salesforce consultancy Elements.cloud, was telling Salesforce Ben that customers “don’t know how to price them yet, and right now, nobody wants to hand out a blank cheque: they want caps and predictable ROI.”

Salesforce’s answer, in June 2026, was Agentforce Help Agent, an autonomous customer-service agent priced entirely on outcomes. Per Salesforce’s own announcement, the company only charges when the agent autonomously resolves an issue from start to finish. If a customer escalates to a human, gives negative feedback, or simply abandons the conversation, there’s no charge at all, and the agent hands the full context to a human rep for free. Salesforce says the model already has a track record: on its own help.salesforce.com support site, the same technology handled 4.3 million inquiries and resolved 70 percent of them without a human. As the company put it in its own announcement, “your cost is tied to outcomes, not activity.”

Flex Credits and the All You Can Eat Contract

Outcome pricing sits alongside two other mechanisms Salesforce introduced earlier. In May 2025, the company rolled out Flex Credits, a consumption currency that lets customers pay for AI usage the way a cloud customer pays for compute: buy a pool of credits, draw them down as agents run, and buy more when the pool empties. Spencer calls topping up that pool “refilling the tank.” Then, in October 2025, Salesforce launched the Agentic Enterprise License Agreement (AELA), an all-you-can-eat contract that caps the price for unlimited AI agent usage over the term of the deal.

Three Pricing Models, Running at the Same Time

That leaves Salesforce selling the same underlying product three different ways at once: the traditional per-seat license, Flex Credit consumption, and outcome-based fees like Help Agent’s pay-per-resolution rate. On top of those three billing mechanisms, Salesforce layers different contract structures, including standard annual commitments, AELAs, and a newer option called Salesforce Commit, which works the way hyperscaler cloud contracts do. A customer might commit to spending $10 million over three years, then split that allowance across seats, Flex Credits, and other consumption as its needs change, according to The Register.

Spencer expects consumption revenue to grow as more customers move AI agents into production, but he was candid about the timeline: it will take another three to five years before it becomes a material share of Salesforce’s revenue, a business that generated $41.5 billion in the fiscal year ended January 2026, according to Salesforce’s own fourth-quarter results.

Why Outcomes Are the Hardest Bet

Of the three models, outcome-based pricing is the one Salesforce itself says is hardest to scale. Patterson acknowledged that some agents “operate across multiple disciplines, domains, and products, making it difficult to identify a single measurable result.” A customer-service agent that resolves a support ticket has an obvious, countable outcome. An agent that drafts a sales proposal, updates a forecast, and flags a compliance risk in the same workflow does not.

Spencer put the underlying problem even more directly to The Register: “The key with outcome-based pricing is you’ve got to be very objective about the measurements that are driving the outcomes, and that always tends to be the challenge.” Both sides, buyer and seller, have to agree in advance on what counts as success, and agree on it in a way that survives a contract renewal two or three years later when the agent’s job has changed.

Gartner’s Early Warning

Salesforce’s own capped AELA contracts have already drawn a formal warning from Gartner. In January 2026, Gartner director analyst Hannah Decker told The Register that unlimited AELA deals are likely to convert into “defined quantity contracts” once the initial term ends, since Salesforce will size the renewal around a customer’s actual usage rather than keep the terms open-ended. “If they’re moving to a defined quantity contract, there needs to be limits on price increases at renewal,” Decker said. In a separate research note, Gartner warned that Salesforce price rises of 6 to 15 percent above inflation would feed directly into renewal uplifts, with genAI and agentic AI product lines pushing those increases even higher.

Decker’s broader point was that customers can’t yet benchmark what they’re being asked to pay: “There is a lot of confusion around how these models work, and that’s not just isolated to Salesforce.” Salesforce pushed back at the time, with Patterson calling the claim that it was moving away from capped agreements “inaccurate” and pointing to a cost-tracking tool called Digital Wallet. The company gave The Register the same assurance again this month: renewals will stay flexible.

The tension Gartner flagged traces back to how Salesforce itself talks about AELA economics. In late 2025, Salesforce’s chief revenue officer, Miguel Milano, told investors he was comfortable losing money on the capped deals in the short term, because a customer who deploys heavily today becomes more valuable later: “The customer deploys so much that all of a sudden, that deal is not profitable for me. And then I have another 20 years to monetize that customer.” That’s a bet on long-term retention that only pays off if Gartner’s predicted renewal shock doesn’t scare customers away first.

A Reliability Problem at the Worst Possible Moment

The pricing debate landed in the middle of a bad week for Salesforce’s own infrastructure. On the second day of Dreamforce, the same event where Salesforce unveiled AIforce and pitched a future built on AI agents handling real work, a login-service failure took down access to Salesforce instances across the US, UK, France, Germany, Japan, and India. Salesforce’s own status updates show the disruption starting at 0830 UTC and the incident finally marked resolved at 1920 UTC, nearly 11 hours later, according to The Register’s live coverage of the event. Salesforce traced the failure to a core system component under increased load, which left requests, in the company’s own words, “stalling while waiting on a response from an internal login service, which is using up available server resources.” The outage also blocked customers from creating new support cases, the very channel that would normally route them to an AI agent for help.

An outcome-based pricing model asks customers to trust that an agent will reliably do its job, or they don’t pay at all. A nearly 11-hour global outage during the exact week Salesforce is asking the market to make that leap is not the reassurance the pitch needed.

What Customers Should Watch For

Not every outside observer thinks the shift toward outcomes is a bad trade for customers, even if the mechanics are still being worked out. Preet Takkar, PwC’s global and US Salesforce practice leader, told The Register he expects outcome-based pricing to become “far more prevalent” by 2030, though “it will take time.” PwC, which has a global alliance with Salesforce, has already introduced outcome-based fees for some of its own clients, putting its own revenue at risk through gain-sharing arrangements “so that our clients do not have to print new money” for AI and digital transformation work, Takkar said.

Until outcome-based pricing matures, the practical advice for Salesforce customers is the same one Gartner has been giving since January: read the exit terms before signing an AELA, ask what happens at renewal if usage has grown, and get explicit caps on any consumption multiplier Salesforce reserves the right to change mid-contract. Decker’s closing line to The Register still applies: “Making sure there are caps that protect you when the agreement ends is critical.”

The Bigger Pattern

Salesforce isn’t the only vendor discovering that agentic AI breaks the licensing math it grew up on. The same tension shows up from the buyer’s side, where Madrona’s own survey data shows enterprises reevaluating AI vendors roughly every six months, undermining the multi-year revenue security that recurring contracts were supposed to provide in the first place. It’s also visible one layer down the stack, where Stripe’s acquisition of the AI model router OpenRouter is a bet that someone needs to build the metering and payment plumbing underneath all this token- and outcome-based billing in the first place. Salesforce has the advantage of controlling both the software and the pricing model. What it doesn’t yet have, by its own CFO’s admission, is a single answer for how much any of it should cost.