Sofy has published the sharpest one-line diagnosis of API testing I have read in a while. In API Test Automation with AI Agents: Solving the Coverage Problem, Syed Hamid writes: “UI testing’s hard problem is maintenance. API testing’s hard problem is coverage.” The arithmetic behind it is the useful part. A single endpoint with five optional query parameters has thirty-two structural combinations before you consider values, authorization states, or error paths, and most suites test the happy path and stop. The advice that follows is to exercise the running API rather than its specification, because “generating tests from the spec verifies that your API matches its documentation. It does not verify that either one is right.”
There are no vendor figures to discount here, which is unusual for a product post. The thirty-two combinations is combinatorics, not a benchmark, and the claim survives on its own. What is worth noting is the design decision the piece is built to justify: Sofy’s agents work against the running API, discovering undocumented parameters, observing real error shapes, and noticing when a field the spec calls required is in practice optional. That is a real position, and it is the opposite of contract-first tooling. The line to keep is the last one: “The cases that break production are the ones nobody imagined, which is precisely why nobody wrote a test for them.”
The catalog lets us turn the argument around and point it at Sofy. The SOFY provider page lists two API pages, both of them the CI/CD surface the catalog description credits the company with. The SOFY Applications API uploads builds, and the SOFY Scheduled Runs API triggers runs, polls status, and aborts them. Error semantics and OpenAPI examples are both lit on the Agent Readiness score, so the real error shapes the post tells you to go and observe are, for Sofy’s own API, described in the contract. The event surface is lit too, which matches the outbound webhooks on run completion.
The Kin Score is 44.9, developing band. Discoverability carries it at 75.9, contract quality at 65.1, developer ergonomics at 58.9. Contract governance is 4.5 and access clarity is 31.6. The Agent Readiness score is 31.5, and the dimension that matters for this story is agentic access, which is unlit. Sofy’s product is agents exercising a running API. The catalog cannot find a description of how an agent would exercise Sofy’s own. The post’s argument is that coverage lives in the running system, not the document, and on that measure the company has covered the document and not yet the door.