Large Language Models (LLMs) like ChatGPT have sparked debate about whether they qualify as intelligent agents. While some argue that LLMs possess agency due to their ability to process information and generate human-like text, others maintain that they are merely sophisticated tools mimicking human conversation. This post delves into the architecture and training of LLMs, revealing their complex inner workings. We explore how LLMs process text, learn from vast datasets, and generate responses. However, a closer examination of their structure reveals their limitations. LLMs lack the fundamental characteristics of autonomous agents, including self-preservation, independent goal setting, and the ability to initiate actions. They react passively to prompts and lack internal motivation. But LLMs are far more than just “stochastic parrots.” They operate within the digital fabric of our lives as linguistic automata, engaging with us in conversations and performing tasks. This unique position introduces new dynamics to human-machine interactions, blurring the lines of agency. We explore the concept of “midtended agency” where LLMs anticipate and complete our thoughts, transforming the nature of creativity and productivity in the digital age. As LLMs become increasingly integrated into our lives, it is crucial to understand the evolving relationship between humans and machines. Are we simply using advanced tools, or are we entering a new era of shared agency where the boundaries between human and artificial intelligence become increasingly indistinct? The future of our relationship with LLMs depends on understanding their capabilities, limitations, and their influence on human agency.