41 Brands Price AI Five Ways. The Credit Is the New Seat.

A chart from AI Momentum Partners is going around LinkedIn this week: 41 software brands, five ways to charge for AI. Misha S., an AI product lead, shared it with a question at the end that deserves a straight answer. As agents reduce the need for human seats, will SaaS charge per user, per unit of work, or per business outcome?

The chart already answers it. The answer is none of the three.

The short version

AI Momentum Partners mapped 41 software brands in October 2026: 21 sell a seat plus a metered AI allowance, 9 bill on usage, 5 bundle AI into the plan, 2 sell a per-seat add-on, and 4 charge per outcome. Outcome pricing gets the coverage, and all four outcome sellers (Fin, HubSpot, Sierra, and Zendesk) price support resolutions. The bigger story is motion: Oracle, Google, Microsoft, GitHub, HubSpot, and Datadog all moved toward credits in the last twelve months, from every column. The PricingSaaS 500 went from 35 companies offering credits at the end of 2024 to 79 a year later. The market didn't pick seats, work, or outcomes. It picked a private currency the vendor mints, prices, and expires. For a product leader, the real price is now three numbers the customer can't see: the exchange rate, the burn per action, and what expires unused.

What the chart actually says

Read the columns as AI Momentum Partners labels them. "Included," AI in the plan with no separate fee: Google, Slack, Shopify, Oracle, Docusign. "Per-seat add-on": Microsoft Copilot and Dropbox Dash. "Seat plus meter," per-person plans plus metered AI usage: 21 brands, from Adobe and Atlassian to Salesforce and Zoom. "Usage-based": 9, including Databricks, Snowflake, Datadog, Replit, and Twilio. "Per outcome": Fin, HubSpot, Sierra, and Zendesk.

The callout box says it plainly. "4 of 41 charge per outcome. Outcome pricing gets the buzz. Seats and meters still do the billing." The chart also marks 18 of the 41 as having documented a change in how they charge for AI since January 2025.

That second number is the one I'd look at. A snapshot of 41 companies, almost half of which changed the model in 21 months, isn't a map of five stable options. It's a photo of traffic.

Every column is moving toward the middle

Here's where each column was heading as of this week, from the vendors' own pricing pages and announcements where they exist.

Two of those deserve a second look.

Datadog was the purest outcome meter in observability: you paid per investigation, and inconclusive ones were not billed. It moved to credits. HubSpot moved toward outcomes, charging only on resolution, and did it inside a credit currency it launched a year earlier. One went toward outcomes and one went away, and both ended up in the same place.

That's the pattern across the board. Bundles are adding meters. Add-ons are adding meters. Meters are turning into credits. Outcomes are being billed in credits. Nobody is moving out of the middle column.

The credit count

The surveys say the same thing from a distance.

Kyle Poyar and Rob Litterst tracked the PricingSaaS 500, the 500 largest SaaS and AI companies with public pricing, through 2025. Thirty-five offered credits at the end of 2024. Seventy-nine did a year later, up 126%. Those 500 companies made more than 1,800 pricing changes in 2025, 3.6 each.

Poyar's 2026 State of B2B Monetization, 230 companies surveyed in April and May, puts hybrid pricing at 37%, up from 25% a year earlier. 29% use AI credits today and another 33% plan to within a year. Three in four changed pricing or packaging in the last year. Asked what they prefer, investors picked hybrid first at 35% and seat-based last at 5%.

Outcome pricing is where the surveys disagree, and the disagreement is informative. ICONIQ's 2026 report has outcome-based pricing at 23% of AI companies. Orb's study of 80 AI agent companies has it at 3.8%, down from 4.5%. The gap is definitional. Count "we have a success fee somewhere in the contract" and you get ICONIQ's number. Count "the invoice is denominated in outcomes" and you get Orb's. The chart agrees with Orb: four of 41.

Gartner's line from July is the one to remember. It sized $234 billion of enterprise application spend as exposed to agents by 2030, and George Brocklehurst put the mechanism in one sentence: "This breaks the link between user growth and revenue growth." Credits are how vendors are rebuilding that link without calling it a seat.

Why credits won

A vendor selling AI is managing two prices that move in opposite directions.

The cost of a token is falling fast. Guido Appenzeller at a16z measured roughly 10x a year at equal capability. The number of tokens a task consumes is rising. Gartner puts agentic tasks at 5 to 30 times the tokens of a chatbot, and ICONIQ's 2026 report has a workflow budgeted at $0.10 a run that went past $1.50 once the agent started retrying. Bessemer's 2025 benchmark had the fastest-growing AI companies at roughly 25% gross margin, often negative. I've written about the first half of this as the Jevons cliff. The second half is what makes a fixed seat price dangerous.

A seat can't absorb either movement. An outcome price absorbs one and exposes you to the other. A credit puts a rate card between the customer and both, and the rate card can change without the contract changing.

Watch it happen. OpenAI cut voice in ChatGPT Business from 5 credits a minute to 1.25 on September 9, a release-notes change. HubSpot halved Customer Agent from 100 credits per conversation to 50, and only on resolution, in April. GitHub moved Copilot to credits consumed per token at model API rates, so the cost of a request now depends on the model and the length of the answer. In each case the sticker price stayed put. The price of the work moved.

That's the shift Misha's question is pointing at, and the chart's five columns hide it. The price left the contract. It lives on the rate card now.

A private currency has three prices

Every credit has an exchange rate, a burn rate, and an expiry date. Only the first one ever shows up in a procurement conversation, and often not even that.

The exchange rate, credits per dollar. Salesforce publishes it: $500 per 100,000 Flex Credits, half a cent each. Microsoft prices Copilot Credits at a cent. Atlassian bills overage at a cent. Datadog lists $1.30 per credit on demand and Snowflake $2 per AI Credit. OpenAI's Business rate card lists credits per action with no dollar rate on the help pages. Workday, SAP, UiPath, and Glean don't publish one either. Without it, a buyer can't compare two vendors, or this year's contract with last year's.

The burn rate, credits per action. This is the number that moves. Atlassian charges 10 credits for a basic Rovo invocation. monday.com lists an agent run at anywhere from 10 to 250 credits. Workday's examples put a resume screen at about 6 and a contract redline at about 500. A burn rate is set by the vendor, published in documentation, and changed in a release. It's the most powerful price in the stack and it has no contract around it.

The expiry, what happens to the credits you don't use. Most pools are use it or lose it. Atlassian's allowance refreshes monthly and doesn't roll over. Neither do Adobe's generative credits. Salesforce's, Workday's, SAP's, and OpenAI Business's expire at the end of the term or after 12 months. In consumer finance this has a name. Breakage is the revenue from gift cards nobody redeems. No AI vendor publishes its breakage, and I couldn't find a credible estimate anywhere.

I have seen what it looks like from the vendor's side. Eighteen months into moving a $40M ARR business from per-seat to hybrid outcome pricing, our tracker had 12 accounts classified as on plan. They were paying their committed minimums. Their real usage sat below the minimum. Nobody escalated, because the dashboard said fine. When the legacy tier sunset, we lost two of them. That's in Drawdown Is Not Adoption, and it's breakage under a different name: revenue that looked healthy precisely because the customer wasn't getting what they paid for.

Now read Salesforce's Q3 fiscal 2026 call with that in mind. CFO Robin Washington: "more than 50% of Agentforce bookings came from existing customers refilling the tanks." Miguel Milano added that 362 customers refilled in the quarter. That's a strong signal, and it cuts both ways. A refill is real usage. It's also the new renewal event, and a renewal is where a customer with a full tank asks why they're buying more.

The outcome column is a support column

All four outcome sellers on the chart price the same thing: a resolved support conversation. Fin at $0.99 per outcome, HubSpot at $0.50, Zendesk from $1.50, and Sierra per completed task.

That's not a coincidence. A resolved ticket is the one AI output that came with a definition customers already counted before AI existed. Help desks have measured resolution for twenty years. Every other outcome has to be defined from scratch, and the definition is where the price breaks. I priced per resolved ticket once. The definition held for 98% of cases and the other 2% became more than 200 contested cases a week and a dispute team that grew from 4 analysts to 14. The numbers are in Field Report: What Broke When We Killed Our Per-Seat Tier.

Even inside support, the definition keeps becoming a product. Zendesk announced at Relate in May that it bills only for resolutions a separate AI evaluation model verifies. Fin now bills several outcome types, from resolutions to handoffs to disqualifications, and sells a sales qualification at $9.99. The single outcome is splitting into a menu, which is the path I traced in Kill the Finish-Line Price.

And then there's the consolidation. Salesforce signed to buy Fin, formerly Intercom, for about $3.6 billion on June 15 and closed on September 10. The best-known per-outcome meter in software now sits inside a company that has run its own agent through a per-conversation price, Flex Credits, per-user add-ons, a flat enterprise license, and, since September, editions with a credit pool per user. Sierra, private and past $100M ARR within 21 months of launch, is the outcome seller still pricing on its own terms.

Outcome pricing isn't dead. It's narrow, it lives where the unit was already defined, and the biggest buyer of it just put it next to a credit pool.

The backlash is the forecast

When the burn rate moves, users find out from the invoice.

Cursor moved its $20 Pro plan from 500 requests to compute-based limits in June 2025, then apologized and refunded unexpected charges a few weeks later. GitHub's move to metered credits on June 1 produced the same week of forum posts. One Pro+ user: "1,180 credits used. 16% of my monthly Pro+ allowance. Gone." Another reported a single request costing more than $6. "Not after a day of usage. After ONE request."

None of those customers was surprised by the price per credit. They were surprised by the burn. James Brown's write-up of Stripe's Monetize conference had the line for it: predictability is becoming a product feature. In a credit world, it's the main one.

Five things the chart gets wrong, or will by next month

The chart is good. It's a snapshot, and parts of it have already moved.

1. Oracle isn't "no extra fee" anymore, per partner write-ups of Release 26C: an included monthly allowance of AI Units, then a meter.
2. Microsoft isn't only a $30 add-on. Copilot Credits meter agent work at a cent each, and E7 bundles Copilot into a $99 suite.
3. HubSpot's $0.50 is billed in credits and only took effect on April 14, 2026. Before that it was $1 per conversation, resolved or not.
4. Amazon Q Business is closed to new customers, per AWS's own documentation. Its replacement, Amazon Quick, prices seats plus agent hours.
5. Fin is now Salesforce. The outcome column has one fewer independent vendor than it shows.

What I'd do with this

If you're setting the price, publish the three numbers before a buyer's CFO asks for them. Put the exchange rate on the pricing page. Show the burn in the product before the action runs, not after, the way a checkout shows shipping before you pay. Roll unused credits over at least one cycle, and if you won't, track breakage internally as a warning light, not as revenue you planned for. Keep one currency. Salesforce has changed how Agentforce is priced at least six times in two years, and every new metric is another thing a renewal conversation has to reconcile. Keep the outcome count next to the credit count in every customer report, the rule from Drawdown Is Not Adoption, so the customer sees what the credits bought.

If you're buying, negotiate the same three numbers. Lock the exchange rate for the term. Get written notice before any burn rate change on the actions you use most. Get rollover, or a quarterly report of what expired.

This sits inside the SaaS to AI Business Models argument: the price follows the work, and the margin follows the price. It also extends Kill the Per-Call Agent Meter, where a meter bolted onto a seat was the problem. The credit is that meter with a currency wrapped around it.

So, to answer Misha's question: not per user, not per unit of work, not per outcome. Per credit, and the fight over the next two years is about who sets the exchange rate. The vendors who publish it will win the trust. The ones who keep it on a release-notes page will win a quarter.

This week: pull your own pricing page and answer three questions in writing. What does one credit cost in dollars? What does your most common action burn? What expires unused at month end, and how much of it did customers pay for last quarter? If you can't answer the third, neither can your customer, and their CFO will ask first.

Related answer: How do SaaS companies charge for AI in 2026?

Sources: AI Momentum Partners, "41 software brands, 5 ways to charge for AI" (October 2026), as shared on LinkedIn by Misha S.; the original report could not be located. Rob Litterst and Kyle Poyar, "The state of SaaS pricing changes in 2025," Growth Unhinged, January 7, 2026. Kyle Poyar, "The state of B2B monetization in 2026," Growth Unhinged, May 13, 2026. ICONIQ State of AI 2026. Orb, 2026 State of AI Agent Pricing. Gartner, $234 billion of application spend at risk from agentic AI, July 1, 2026. Gartner on inference cost and agentic token use, March 25, 2026. Guido Appenzeller, "LLMflation," a16z. Bessemer, The State of AI 2025. Salesforce Q3 FY26 earnings call transcript. Salesforce signs definitive agreement to acquire Fin, June 15, 2026, and completes it, September 10, 2026. Agentforce's pricing journey, Pricing Innovation. HubSpot on Customer Agent outcome pricing. Atlassian Rovo licensing. GitHub Copilot moves to usage-based billing. The Register on GitHub Copilot metered billing, June 2, 2026. TechCrunch on Cursor's pricing apology, July 7, 2025. Datadog pricing and PricingSaaS on the Bits AI SRE change. Oracle AI Agent Studio launch and Version 1 on 26C pricing. Google Workspace price change. Slack June 2025 pricing. Microsoft 365 Copilot pricing. OpenAI ChatGPT Business release notes. Amazon Q Business availability change. TechCrunch on Sierra at $100M ARR. The dispute and drawdown figures are from Field Report: What Broke When We Killed Our Per-Seat Tier and Drawdown Is Not Adoption, falkster.com.