Notion's Agent Built Its Own Integrations, Stripe Explains AI Pricing

Stripe's PM Revenue lead put a name to the pricing problem every AI PM is wrestling with right now ("alien economics"), while Cursor keeps shipping agent infrastructure (canvases, an app marketplace) faster than most teams can evaluate it, and Karri Saarinen is asking the question underneath both: when agents do the building, who's actually making the product decisions? ### Stripe - The "Alien" Economics of AI Monetization **Source:** https://stripe.com/blog/industry **Credibility:** High (first-party, authored by Stripe's Product Manager for Revenue) **What happened:** James Brown, Stripe's PM for Revenue, published a piece on October 13 arguing that AI companies are hitting a pricing model that doesn't behave like traditional SaaS. The core claim: usage costs, customer expectations, and value delivery are all moving simultaneously and unpredictably, which breaks the assumptions baked into standard per-seat or flat-tier pricing. Stripe frames this as a structural problem, not a temporary one, which matters coming from the company that processes pricing changes for a huge share of the AI economy. **Key technical details:** - The piece is framed around three moving variables: usage costs (compute, inference), customer expectations (what "normal" feels like keeps shifting as models improve), and realized value (output quality varies per request, unlike a fixed SaaS feature) - Stripe positions this as different in kind from prior platform shifts (mobile, cloud) because the cost-to-serve is volatile at the unit level, not just at the aggregate business level - Comes paired with other Stripe product news this week: new money movement and cross-border tooling for fintechs (Oct 16, led by Neetika Bansal) **Why it matters for PMs:** If you're pricing an AI feature, this validates what probably already feels true: the flat-tier-plus-overages model that worked for SaaS doesn't map cleanly onto a cost structure where your COGS per user action swings based on model choice, prompt length, and context window. Stripe is signaling (via a PM, not a marketer) that this is worth building new tooling around, not just negotiating around. **Critical questions:** - Is Stripe previewing a product response here (dynamic/usage-based billing tooling), or just naming the problem? - How do you communicate volatile unit economics to customers without looking like you can't control your own costs? - Does this argue for absorbing margin volatility yourself (flat pricing, worse margins) versus passing it through (usage pricing, worse UX)? **Action you could take today:** Pull your last 90 days of per-user AI feature cost data and check how much it varies user-to-user. If the variance is high, your current pricing model is probably mispriced for your heaviest users, your lightest users, or both.