Pricing in the AI era

LAST REVIEWED 2026-08 · SOURCED FROM 2 SESSIONS, APR–MAY 2026

TL;DR

The rule of thumb from the room: if the end consumer of your product is a human, seats survive; if it’s an agent, sell outcomes or usage. Anchor any price against the dollars the buyer already spends on the problem. Keep usage revenue out of your ARR line when you report to investors. And for early deals, the price is less important than what the discount buys you — proof points are worth more than margin at this stage.

Seats for humans, outcomes for agents

The heuristic that organizes everything else, from the session on software’s repricing:

End consumer	Pricing model that survives
A human logging in	Seats still work
An agent doing the work	Outcome or usage pricing

The examples cited in the room: a major CRM vendor pricing its AI assistant per conversation (about $2), support platforms pricing per resolved ticket, and app-building tools selling usage top-ups the moment the customer’s agent needs more capacity. The common thread is that the unit of value shifted from “a person with access” to “a job completed” — and pricing followed.

Two implications for a startup picking a model today:

• Don’t price a system of action like a system of record. If your product does the work, per-seat pricing understates your value and invites the comparison to cheap seat-based incumbents.
• Expect hybrid stacks. A platform fee (predictability for the buyer) plus usage or outcome pricing (upside for you) came up repeatedly as the emerging default.

Anchor against existing spend, not your costs

From the office-hours session on first deals: price discovery starts with the buyer’s current cost of the problem. “You’re already spending $500K a year on this” reframes a $50K contract as a 90% saving instead of a new expense. If discovery hasn’t produced that number, keep discovering — the price conversation is premature.

The corollary discipline for AI-native products: know your variable cost per unit of work (tokens, compute, human review) so outcome pricing doesn’t quietly invert your margin at scale. Investors will ask exactly this — see the unit-economics expectations in Data rooms and diligence readiness.

Reporting it honestly

Two rules repeated across the finance and legal sessions:

1. Usage revenue is not ARR. Keep license/subscription ARR separate from variable usage revenue when you report. Blending them is the kind of thing that unravels in diligence.
2. Explain your margin. A 70%+ gross margin is the SaaS default; an AI-native ~30% margin is acceptable when the variable-cost story is clear and the path up is credible.

Early-stage pricing is really proof-point purchasing

At the first-customers stage, the biggest pricing mistake is optimizing the number instead of the structure. The full playbook is in First customer contracts: full price with a visible discount, the discount traded for utilization, feedback, KPIs, and a contracted case study — and lowballs countered on duration, not price.

Sources

Two sessions: a talk by a multi-exit founder on AI-driven pricing-model shifts (Apr 2026), and fundraising office hours with an exited founder covering price discovery and first-contract structure (May 2026), with reporting rules from the finance session (Dec 2025).