Get the latest tech news
GPT-5 vs. Sonnet: Complex Agentic Coding
OpenAI released GPT-5 yesterday, promoting it as their best model yet for agentic coding.
After babysitting the agent, restarting it a couple of times, and twice confirming it should continue with the implementation, it managed to create a complete port that was provably working functionally the same as the original. Unlike GPT-5's single messy Rust file, Claude created a very neat project structure with different modules for each part of the program — much more readable and maintainable code I would consider working with. One persistent problem: Even though Copilot Chat lets you allow actions initiated by tools and MCP servers for the session duration or for all activity in the workspace, terminal commands often require manual approval.
Or read this on Hacker News
