What I Learned from Working with AI Over the Past Few Years

Summary

Simon Späti, who has published writing on his blog since 2015, reflects on using AI over the past two to three years while continuing to write manually. He argues that writing intended for people should generally retain a human author, whereas code can be generated more readily because it primarily serves as a tool for computers, subject to team and maintainability concerns. Späti says AI-suggested ideas are difficult to complete because they lack his own interest and connection, so he prefers manual writing, daydreaming, and following his own train of thought. He uses Claude Code extensively for Omarchy plugins and Linux optimization or failure detection, but not for writing or note-taking. He also argues that people who began writing only after AI became widespread may find it harder to demonstrate what they can produce without AI, just as newer products can be difficult to assess when every product appears impressive. In his view, AI cannot yet replace the human thinking involved in writing because it does not write from personal experience or the heart, even when its prose is technically good. He distinguishes writing for humans from writing for machines and agents, while suggesting that concise, clear documentation can serve both. He concludes that readers should be told whether text was AI-written because AI assistance can change the text’s intent and development.