Why Cursor, Attio, and Linear Still Charge per Seat for AI Products

A leaked copy of the S-1 Anthropic confidentially submitted June 1 shows 2025 revenue of $4.59 billion up 1,088% from the year before. The operating loss was $8.06 billion, and the company has $20 billion cash balance. Anthropic also shipped Claude Sonnet 5.5, which it says costs up to 30% less per task at an unchanged $2 per million input tokens and $10 per million output. Microsoft moved Copilot to per-user seat plus usage-based agent features. Frontier models sit on a separate line. And Sierra's Bret Taylor posted about Ghostwriter running proactive @-mentions on issues no one asked it to check.

All four point the same way. Seat-based AI contracts are losing ground to consumption pricing. We've made that case three times in five days, from Thursday's piece on Bret Taylor's HumanX keynote to Monday's on Nebius's price hike, announced September 17, and OpenAI's $500 Pro Max tier. The wave is real, and it carries the risk you'd expect from a company that sells tokens instead of seats: when the biggest buyers run less, revenue moves with them.

What decides AI pricing is who owns the outcome. How AI-native the company is matters far less.

When the agent owns the work end-to-end, consumption pricing wins. Ghostwriter flagging an issue before a human asks is agent-owned work. So are the autonomous features Copilot now meters on their own line. When the human still creates the value and the AI accelerates them, per-seat still wins. Three AI-native companies have now bet on that second case on purpose.

Three AI-native companies kept the seat on purpose

Cursor's users love per-seat pricing. That's a direct quote from Cursor's own June 2025 pricing post, said again in its June 2026 Teams Premium launch. Their words: “our users love per-seat pricing, it's just the cost side makes it harder.” Cursor is the definitional AI-native product. Its model costs would normally push it toward consumption pricing, and its users still prefer a seat. The customer wanted the predictable line item, and Cursor doubled down 12 months later with a hybrid tier that keeps the seat and adds a metered pool on top for the parts it can't absorb.

Attio runs the same play more cleanly. Its pricing page sells seats at $29 to $79 each, with a separate pool of Seat Credits attached to each user for AI features the user triggers. Automations that run without anyone watching draw on a separate Workspace Credit pool. Attio raised seat prices in July 2026 and kept the structure. The sales rep is still a seat because the sales rep is still creating the value. The automation is its own meter because it runs on its own.

Linear makes the clearest case. It has two pricing pages for the same product, split precisely along the frame's axis. Human planning work runs $10 to $16 per seat. Coding Sessions, the autonomous work, run about $5 to $7 per session on a separate line. Every seat gets $20 in promotional credits to seed usage. Karri Saarinen's Self-Driving SaaS essay from October 2025 states the mechanism directly: real change comes when AI stops requiring constant input and instead proactively moves work forward on its own. Linear read the same wave Anthropic and Microsoft are riding and priced the two halves of its product accordingly rather than picking a side.

Notion's CEO Ivan Zhao is the strongest voice on the other side. In 2025 interviews he said out loud what most AI-heavy founders think but rarely publish. “You're no longer just selling seats, you should think about how you sell usage, sell work, sell outcomes.” Zhao is right about where the market is heading. What Cursor, Attio, and Linear are demonstrating is that getting there won’t leave a single pricing model standing. The market is splitting in two.

The counter-case worth taking seriously

The strongest counterargument is that these AI-native holdouts are pricing this way because they haven't figured out consumption pricing yet. On this view, rising costs will force them to switch. Selling on consumption at scale ties up a lot of the seller’s capital, and Cursor's own quote admits as much. Seen this way, per-seat AI-native pricing is a leftover from the pre-scale phase.

However, the reading breaks on Linear. Linear already sells consumption pricing. It charges the same customer both ways in one product, and who does the work decides the meter. That's a choice driven by the outcome, and cost comes second. When the vendor closest to autonomous work ships two pricing pages on purpose, “switch to consumption or die” looks like the wrong lesson.

On October 5, Vanta CEO Christina Cacioppo wrote in a LinkedIn post that token usage in Vanta is up 30x since January and that Vanta doesn't charge for tokens. The 30x is company-reported, with no baseline or cost figure attached, and she called tokens “one measure of how much tedium the agent is absorbing.” That's the cost pressure the counterargument predicts, and it lands on work Cacioppo credits to the agent. Vanta has so far kept that cost inside its plan pricing rather than add the kind of meter Microsoft and HG Insights attached to the seat.

What this means for CROs

If you are a CRO reading this week's pricing news, don't rebuild your pricing page around the strongest signal of the week. Start from who owns each outcome in your product. If a human is still the creator and the AI is the accelerator, the seat has more life than Bret Taylor's HumanX keynote suggested. If the agent is running work with no human in the loop, the meter is honest and the seat is a lag. The pricing page most likely to survive the next two years has both, split the way Linear’s already is.

Sources:

• Anthropic's June 1 confidential S-1 submission; leaked S-1 details (SiliconANGLE)
• Anthropic Claude Sonnet 5.5
• Cursor’s June 2025 pricing post
• Cursor’s June 2026 Teams Premium launch
• Attio pricing
• Linear: Self-Driving SaaS

What decides AI pricing is who owns the outcome. How AI-native the company is matters far less.

Before changing your pricing page, mark who owns each outcome in your product. Keep the seat where a person creates the value, and meter the work an agent runs on its own, the way Linear bills Coding Sessions on a separate line.

By Business of GTM