TL;DR: Salesforce has priced one AI product four ways since May 2025, and the market splits between charging for effort (tokens), output, and outcomes (all-you-can-eat wasn't sustainable after all). The vendors coping best sell a currency instead: credits whose exchange rate they control, so the next change happens in a rate table. Fund usage data and the currency layer; defer billing, contracts, and comp tied to one unit. Your customer's budget, built on your unit, is what you can never migrate. Output or outcome pricing leaves rising inference costs on your P&L, and not in a good way. Freeze the exchange rate for each contract term so customers can budget. Bill this year's usage through two candidate units, on paper, and record what breaks.
Want to listen to this article?
Subscribers to the Orion Playbook newsletter can listen to the AI-Generated Audio version of this article for free. Already a subscriber? Log in here
In May 2025, Salesforce moved Agentforce off $2 a conversation and onto credits per action, with per-user licenses beside them. In July 2026, it added a fourth price: by the resolution. This company has arguably the deepest pricing bench in software. If Salesforce has not settled what to charge for in AI, a mid-market company that says it has picked its business model is stating a hypothesis.
The rest of the market is no more settled. Bain looked at roughly 200 B2B software vendors in June: 35% charge for effort (tokens, agent hours), 55% for output (drafts written, records updated), and 10% for outcomes. Cursor found out what a wrong unit costs. In June 2025, it swapped a monthly allowance of requests for a dollar budget at API prices. Less than three weeks later, its CEO apologized and refunded the surprise charges, explaining that "the hardest requests cost an order of magnitude more than simple ones." A request stopped being a fair unit once the models started thinking longer.
The vendors coping best stopped trying to pick the unit. They sell a currency instead. Bain found that about four in five vendors adding AI pricing now sell capacity: credits or prepaid commitments, with overages on top. Atlassian, ServiceNow, and Adobe all sell a version of it. Salesforce went furthest, letting customers convert user licenses into credits and back. The customer buys the currency. The vendor owns the exchange rate: how many credits an action costs, and which actions exist. When the right unit changes, the vendor edits a rate table and no contract reopens. Your customer thinks they bought a price. They bought an FX rate, and you set it.
Per-user seats remain, because a seat records which person is authorized to act (the case I made in The SaaS Bill Just Split Into Two Meters). What keeps moving is the price of the agent work done under that authority, and that is where the currency belongs.
So readiness means keeping what the customer buys apart from what you count. Usage data captured at the event level survives any unit, and so does the currency layer. Fund both. Billing rules keyed to one unit, contract language that defines it, and comp plans that pay on it are bets. Defer them. The costliest bet sits outside your walls: the customer's budget, built on your unit, in a spreadsheet you cannot migrate. When we moved CorelDRAW from perpetual licenses to subscription, the transition took a little over three years from the moment we launched our SaaS offering, and we at least knew the destination. The new price was the easy part. The internal tooling and workflows took some effort. Transitioning the customers who had budgeted for the old model was the hard part.
The unit also decides who holds the inference bill. Per-token prices keep falling. Gartner still forecasts that inference cost per agentic workflow will rise more than fivefold through 2028, because each model generation spends more tokens to finish a task. Charge for effort, and that rise passes to the customer. Charge for outputs or outcomes, as about two-thirds of the market does, and it impacts your P&L. Salesforce's per-resolution offer says so in writing: the platform is "unmetered during the agent interaction." A fixed price per outcome inside a multi-year contract is a short position on your own cost base, sold before the next model ships. The currency is how you move that line: repricing an action in credits is a rate-table edit, available at every renewal.
Now sit in your customer's chair. The exchange rate you keep is their budget variance. This newsletter is part of an experiment in running a one-person business with AI agents, and the tokens one piece of work consumes shift with every new model and every software harness update. My forecast breaks without anyone touching a price. Hold the rate for the contract term and move it only at renewal, with notice. You keep the lever; they keep a number they can defend to their CFO. The objection still stands. Credits are opaque, and procurement will push to lock the rate for years, which binds the unit all over again. For a $100M company without Salesforce's systems team, committing now and rewriting later may cost less than a buffer it never uses. The test is your cost per task. Commit to one unit when it held steady through the last model upgrade and your biggest customers budget once a year. Keep the currency when a single model release can move it inside one contract term. If yours sits in between, get in touch, and we can work through your specific situation.
Run a shadow invoice. Take this year's actual usage on one product line and bill it, on paper, through the two pricing units you would most likely move to. The first finding is usually that your usage data cannot produce the bill. Date that finding; it is your honest answer to the question at the top of this article. Then fund only the work that survives any unit in next year's plan, and name an owner for the rate table, a seat between finance and product that most mid-market companies have not filled. A CEO should be able to answer the readiness question in one sentence at the next board meeting.
The price list is moving out of the contract and into the product. Whoever owns the rate table now owns your price.
All the images were generated with AI (ChatGPT Images, Gemini Nano Banana, Claude Opus) by Gérard Métrailler.
Sources
Gartner. "Gartner Predicts AI Inference Costs Per Agentic Workflow Will Increase More Than Fivefold Through 2028." Press release, August 17, 2026. https://www.gartner.com/en/newsroom/press-releases/2026-08-17-gartner-predicts-ai-inference-costs-per-agentic-workflow-will-increase-more-than-fivefold-through-2028. Accessed 2026-09-27.
Maltiel, Eric, and Josh Sandberg. "AI Pricing: A Reality Check on Effort, Usage, and Outcomes." Bain & Company, August 10, 2026. https://www.bain.com/insights/ai-pricing-a-reality-check-on-effort-usage-and-outcomes/. Accessed 2026-09-27.
Salesforce. "Salesforce Introduces New Flexible Agentforce Pricing to Accelerate the Digital Labor Revolution." Press release, May 15, 2025. https://www.salesforce.com/news/press-releases/2025/05/15/agentforce-flexible-pricing-news/. Accessed 2026-09-27.
Salesforce. "Salesforce Launches Agentforce Help Agent That Deploys in Minutes and Only Charges for Resolutions." June 25, 2026. https://www.salesforce.com/news/stories/agentforce-help-agent-announcement/. Accessed 2026-09-27.
Truell, Michael. "Clarifying our pricing." Cursor, July 4, 2025. https://cursor.com/blog/june-2025-pricing. Accessed 2026-09-27.
