Thomas Ptacek treats LLMs as adversarial copyeditors

Relying on LLMs to write drafts produces bland, homogenized slop, but using them as adversarial, rule-bound copyeditors is a high-leverage workflow that sharpens human prose without sacrificing authorial voice.

• –Enforcing a strict ban on adopting LLM-suggested wording acts as essential cognitive defense against the glossy, magazine-headline cadence typical of frontier models.
• –LLM sycophancy actively undermines revisions by validating raw first drafts, making it essential to strip out encouragement and blind-test rewritten passages without conversational context.
• –Grounding copyediting prompts in formal style frameworks turns mechanical editing into a deterministic, bug-hunting process that models execute without fatigue.
• –Personal style lives in the friction of rewriting; models should only surface structural weaknesses, leaving the resolution entirely to the human author.