Microsoft found many of the world's most popular AI tools, such as those from OpenAI and Anthropic, were generally ineffective at debugging code, even with significant third-party assistance.