SaaStr: How AI agents are rewriting the economics of the software stack

Jason Lemkin, founder of SaaStr, published this analysis in June 2026 after noticing a contradiction in his company’s software bills. SaaStr runs with three humans and over twenty AI agents, and the economics of its SaaS stack had fractured sharply along a single fault line: does an AI agent need the software to do its job?

The Salesforce case is the clearest example. SaaStr reduced its human Salesforce seats from more than ten to two. Despite this, the annual bill climbed 83%, from roughly $12,000 to $22,000. The reason: AI agents access Salesforce at roughly 100 times the rate humans did, treating it as a central operational hub through API calls and consumption-based pricing. Fewer humans, far more usage, higher bill.

Notion tells the opposite story. The company had used Notion extensively. After deploying AI agents, the team stopped using it entirely. Their agents built dashboards and interfaces elsewhere. Notion’s human-oriented design had no surface area for the agents to attach to, so when the agents stopped needing it, the humans followed.

Lemkin distills this into a structural question for every B2B software company: do AI agents require your product to be successful at their jobs? If yes, the platform likely sees higher usage and stronger retention as AI adoption accelerates. If no, the platform risks displacement — not by a competitor product, but by custom automation the agents build on their own or by different tools they prefer.

For product managers building B2B tools, this case introduces questions that did not exist two years ago. How do AI agents interact with the product? Can they call the API directly, or are they stuck behind a UI designed for humans? What happens to pricing and retention when the agent-to-human ratio at a customer account inverts? A product team focused on daily active users and session length may miss the shift entirely while it is happening.

The SaaStr example does not generalize automatically. Different product categories have different exposure to agent substitution. But the framework it offers — “do agents need this to succeed?” — is one that product teams building for B2B contexts should run against their own roadmap. The answers will not always be comfortable, which is probably why most teams have not asked the question yet.

This article is useful for product managers working on B2B SaaS products, for founders evaluating the durability of their go-to-market model, and for anyone responsible for pricing strategy as AI adoption inside customer organizations accelerates.