HubSpot Lost 21% Because Its AI Agents Actually Work

HubSpot beat Q2 revenue at $912M, up 20%. Beat EPS at $3.26 against $3.02 expected. Its Data Agent is at 16,000 customers, up 80% quarter over quarter. Its Customer Agent resolves 72% of tickets without escalation.

The stock had its largest one-day drop in 12 years on the NYSE, down roughly 21 to 23%.

Net customer adds came in at 7,000 against 9,000 to 10,000 guidance, and Q3 guidance got cut to 5,000 to 6,000. Net revenue retention fell to 102%, and the reason is the part that should make every SaaS founder sit up: not churn, downgrades. Management said out loud that customers are asking to swap seats for credits.

That's the whole story in one sentence. Every ticket the Customer Agent resolves is a support seat nobody renews. The product works, and the working product eats the pricing model.

The cross-category corroboration is what makes this more than one bad quarter. Salesforce is running at least three concurrent Agentforce pricing models right now: Flex Credits at roughly $0.10 per action, $2-per-conversation, and $125 to $550 per user per month bundles, plus a pay-per-resolution Help Agent that went GA in July with PenFed, Fisher & Paykel and PowerSchool live. Three models simultaneously isn't strategy, it's a company that doesn't know the answer hedging in public. Zendesk moved its AI agents to resolution-based pricing entirely. Shopify's usage-based fees climbed to 21% of subscription revenue versus 11% in Q1 2023.

And here's the number that explains the timing problem: Agentforce is deployed by roughly 12% of Salesforce's 150,000+ customers at $540M ARR. The pricing model is changing years ahead of the adoption curve. Revenue math looks terrible in the gap between "we stopped selling seats" and "consumption revenue caught up."

If you're building anything with per-seat pricing and an AI feature that reduces human work, you're on this curve. I'd rather price for the transition now than get repriced by customers mid-renewal. The specific lesson from HubSpot isn't "don't build agents." It's that the pricing conversation with your customers starts the day your agent starts working, not the day you're ready.