Simon Willison shared his predictions for the tech industry in a recent podcast episode. He expressed uncertainty about the future, particularly due to rapid advancements in coding agents. In the coming year, he believes it will become clear that large language models (LLMs) can write effective code, challenging the prevailing skepticism among some programmers. He attributes this shift to the development of reasoning models and reinforcement learning techniques that have significantly improved code quality. Willison anticipates that 2026 will also see breakthroughs in sandboxing, allowing users to run potentially harmful code safely. He pointed out the current risks of executing code from unknown sources and highlighted the need for better user experience in existing technologies like containers and WebAssembly. He also predicts a significant security incident related to coding agents, drawing parallels to the Challenger disaster, where complacency can lead to serious consequences. Looking ahead to three years from now, Willison raised two key points. First, he referenced the Jevons paradox, questioning whether advancements in coding agents will devalue software engineering jobs or create greater demand for custom software. Second, he forecasted the development of a new web browser primarily built using AI-assisted coding, countering skepticism about LLMs' capabilities for large projects. He believes that within three years, the effectiveness of AI in serious software development will be widely accepted.