This week saw major developments in AI coding tools. Google launched agent mode for Gemini Code Assist, allowing AI to act as a pair programmer that understands entire codebases and makes multi-file changes.

However, a surprising study found AI tools like Claude 3.5 actually slowed developers down by 19%. Developers thought they worked faster, but time was lost reviewing and fixing AI suggestions.

AI coding tools are shifting from editors to terminal interfaces. Companies like Anthropic and OpenAI released command-line tools that handle project setup and dependencies better than traditional editors.

MIT researchers highlighted challenges beyond code generation. AI struggles with refactoring, bug fixes, and system migrations. They called for better data and tools to make AI true engineering partners.

In Poland, programmer Przemysław "Psyho" Dębiak beat OpenAI's custom AI in a 10-hour coding competition. He said the AI pushed him to perform better, showing humans still excel in complex problem-solving.