Summary

Ben from Hugging Face explores the potential of coding agents in AI systems engineering, highlighting their ability to solve complex technical challenges like writing optimized CUDA kernels. The talk breaks down increasingly complex tasks, from interactive agent-assisted coding to zero-shot model training and multi-agent research setups, demonstrating the growing acceptance and capabilities of AI coding assistants. The key takeaway is that engineers should embrace these tools to tackle more advanced problems, moving closer to the technological core and continuously challenging themselves in emerging areas of AI development.