When chat is the wrong UI (github.blog)

A recent discussion highlights the limitations of chat as the primary user interface for interacting with AI, particularly large language models (LLMs). While chat has become popular due to its accessibility, it may not always be the most effective way to leverage AI capabilities. The argument, supported by academic Steven Pinker, suggests that a more task-oriented interface could unlock greater potential. The GitHub Copilot app introduces the concept of a "canvas" — an interactive, customizable UI that allows users to automate their workflows and engage with AI more meaningfully. Unlike traditional chat, canvases can execute code, manage local applications, and communicate bi-directionally with the AI agent, enhancing user experience and efficiency. The significance of this shift is immense for the AI/ML community, as it suggests a move towards more specialized applications of AI rather than one-size-fits-all solutions. With full-stack functionalities, canvases can provide intuitive UIs for various tasks, such as game-building or database management, demonstrating their versatility. This approach encourages developers to rethink their interaction with AI, moving away from token-consuming chat sessions to more productive and effective automated processes. Ultimately, adopting customizable UIs like canvases could greatly enhance workflow automation and innovation in utilizing AI technologies.