Simon Willison presented a lightning talk at PyCon US 2026, summarizing significant developments in large language models (LLMs) over the past six months, highlighting the November 2025 inflection point in coding agents' capabilities and the emergence of powerful local models that exceed expectations. Key themes included the rapid improvement of coding agents and the competitive landscape among major LLM providers.
The key insight for you is that coding agents have dramatically improved over the past six months, with OpenAI and Anthropic enhancing their models through Reinforcement Learning from Verifiable Rewards, making them reliable daily drivers for coding tasks. This development, alongside the emergence of powerful local models, indicates a significant shift in AI coding capabilities, offering new opportunities for integrating and leveraging AI in professional workflows.