Pricing is the most under-thought decision in early AI companies. With inference costs that scale per request and value that often replaces labor rather than augmenting it, the old SaaS seat model frequently breaks. Here's how we think about it.

Seat-based pricing is dying for agents

If your product replaces work rather than helping a person do work, charging per seat caps your upside and misaligns with the value you create. Agents that complete tasks autonomously have no natural 'seat' to attach to.

Usage-based protects your margins

Because inference is a real marginal cost, usage-based pricing keeps gross margins healthy as customers scale. The risk is unpredictable bills — so the best companies pair usage pricing with committed-spend tiers that give customers budget certainty.

Outcome-based is the prize

Charging for outcomes — resolved tickets, qualified leads, closed books — aligns price perfectly with value and commands the highest willingness to pay. It's hard to measure and attribute, but the companies that pull it off build the deepest moats.