Why Your Coding Agent Can't Be Your Testing Agent

Coding agents that write code are structurally unsuited to test that same code because they verify their own assumptions rather than challenging them. The core argument is that testing requires the absence of the author's mental model — the same reason we don't ask authors to grade their own exams. A dedicated testing agent like DevAssure O2 approaches PRs cold, thinks in user personas rather than endpoints, actually runs code in a real browser, and self-maintains as the UI evolves. Three failure modes of self-testing are outlined: tests that mirror bugs in the code, false confidence from green checkmarks on untested user journeys, and a maintenance treadmill that turns automation into theater. The conclusion is that coding agents and testing agents serve fundamentally opposite goals and should be separate tools.