The journey from Clippy to artificial intelligence has been gradual, with digital assistants evolving from hardcoded agents using the Understand, Decide, Respond model to more advanced AI models like Natural Language Understanding and Large Language Models. These models enable agents to predict their responses and actions based on a confidence-ranked range of possibilities and given inputs, producing more fluid and natural conversational experiences. However, the risk of "hallucinations" remains, particularly in tasks that require high accuracy, such as issuing refunds. Recent advancements in Large Action Models (LAMs) have the potential to make agents fully probabilistic end-to-end, but their effectiveness and reliability are still uncertain. As LAMs become widespread, they may change how agents are built, with decision logic being abstracted and simplified, and conversation designers focusing on designing permission systems for where and when a LAM can call a particular function. The adoption of LAMs is expected to bring about new opportunities and challenges for the industry, including the need for more advanced training and management tools.