Engineer Shares Doubts on AI Agents Changing Tests

Dax Raad posted that good tests rarely change but agents constantly alter tests during their work. He described this as agents basically writing the same code twice. Raad noted he had not sat down to think deeply about the issue yet something smells weird to him here. He concluded the observation probably is not worth much. The comment comes from the software engineer known for the SST framework who is currently building an AI coding agent called OpenCode.ai.