Outcome-based pricing for AI startups

Pricing is a go-to-market decision, not a finance afterthought. For AI products, it is also a strategy decision. Sequoia Capital has argued across three essays that AI companies should charge for completed work rather than for access. This guide turns that argument into practical choices.

Seats versus outcomes

In "Generative AI's Act o1" (October 2024), Sequoia contrasted two eras. Cloud companies targeted the software budget and sold per seat. AI companies target the services budget and sell per outcome. The essay uses Sierra as its example. Businesses put Sierra's agent on their website to resolve customer issues, and Sierra is paid per resolution. There is no seat.

The reason is structural. When an agent does the work, the number of human users can fall as value rises. A per-seat price then shrinks as the product succeeds.

Copilot first, autopilot second

Sequoia describes a common path. Deploy as a copilot with a human in the loop, then use that experience to earn the right to run as an autopilot. The Act o1 essay cites GitHub Copilot as a copilot example and notes that Sierra benefits from a graceful failure mode: escalation to a human agent.

Julien Bek's March 2026 essay sharpens the distinction. A copilot sells the tool to a professional. An autopilot sells the work to the end customer. Bek argues that the starting position matters because it determines which customers you can win now.

Choosing the outcome unit

The unit must be something the customer can count and agree was delivered. A good unit has a clear finish line and a quality check both sides accept. It should also map to a budget line the customer already has. Bek's examples include a drafted NDA and a closed month of books. In customer support, the unit is a resolved ticket.

Bek recommends starting with tasks the customer already outsources. The customer already buys an outcome, so the unit and budget exist.

Pricing questions from Sequoia's 2026 essay

In "2026, This is AGI," Pat Grady and Sonya Huang ask founders whether they can price and package to value and outcomes. They pair that with a reliability question: whether the founder is improving the agent harness and has a strong feedback loop. The two are linked. Outcome pricing only works if the product delivers outcomes consistently.

Risks to manage

• Disputed outcomes. Define success in the contract before launch, including how disputes are settled.
• Cost variance. Some tasks use far more compute than others. Track cost per outcome by task type.
• Revenue predictability. Per-outcome revenue moves with volume. Consider a monthly minimum with per-outcome pricing above it.
• Failure handling. Decide who pays when the agent escalates to a human.

Founder takeaways

• Pick one unit and test it with five customers. Use their language for the outcome.
• Anchor the price to the current cost of the work. Compare it to the agency invoice or staff time it replaces.
• Measure gross margin per outcome weekly. Margin should rise as models get cheaper.
• Start with a human in the loop where errors are costly. Move to autopilot as accuracy data builds.

PacTech's Go-to-Market and Sales module includes a pricing workshop where founders test outcome units with real prospects.

Sources

1. Generative AI's Act o1
   Sequoia Capital (Sonya Huang, Pat Grady) · Oct 9, 2024
2. 2026, This is AGI
   Sequoia Capital (Pat Grady, Sonya Huang) · Jan 14, 2026
3. Services: The New Software
   Sequoia Capital (Julien Bek) · Mar 5, 2026