Writing is thinking. LLMs generate the next token via autoregressive decoding — a probability distribution over language, sampled one word at a time. The more we delegate this to machines, the better they get. But if we stop writing, we stop thinking. And lose the very minds that built them.