Jason Lemkin launches API grader
- Jason Lemkin, founder of SaaStr, said this week he launched an “Agentic API Grader” to score how well software APIs work for autonomous AI agents, not just human developers. - Lemkin tied the project to real examples, saying Stripe scored A-95 while HubSpot “gave us a zero,” framing the grader around whether agents can reliably discover, call, and complete tasks. - The launch lands as Stripe and OpenAI push the Agentic Commerce Protocol, a new beta standard for agent-to-business transactions. (github.com)
Jason Lemkin says he has launched an “Agentic API Grader” to rate how usable software APIs are for AI agents, rather than for human developers alone. (saastr.com) Lemkin described the project in posts and related SaaStr writing this week while discussing how his team is testing dozens of agent workflows in production. He tied the grader to concrete examples, including Stripe at A-95 and HubSpot at “zero.” (x.com) (saastr.com) The premise is simple: an API that looks fine in docs can still fail an agent if permissions are unclear, actions are hard to discover, or responses are too inconsistent for automation. Lemkin’s framing shifts the test from “can a developer wire this up?” to “can an agent finish the job?” (saastr.com) That question has become more urgent as software vendors rush to make products callable by large language model systems that plan, choose tools, and execute multi-step tasks. In that setup, the API is less like a menu for engineers and more like a control surface for a machine operator. (aws.amazon.com) (azure.microsoft.com) Stripe is one of the clearer examples of the shift. Its developer docs now include an agent toolkit that works with OpenAI’s Agents SDK, Vercel AI SDK, LangChain, and CrewAI, and Stripe tells users to run those systems in sandbox mode and limit access with restricted keys. (docs.stripe.com) Stripe is also working with OpenAI on the Agentic Commerce Protocol, an open specification now in beta for connecting buyers, businesses, and AI agents to complete purchases. The repository shows a release dated April 17, 2026. (github.com) Lemkin has been arguing more broadly that software companies now need “agentic deployment” skills, not just chatbot features. In recent SaaStr posts, he has described AI agents as a new operating layer for sales, marketing, and support workflows. (saastr.com 1) (saastr.com 2) The grader turns that argument into a scoreboard. If more vendors start publishing agent-ready interfaces, the next API competition may be less about elegant docs and more about whether an autonomous system can actually complete a task without breaking. (saastr.com)
