a16z crypto says the real problem with AI writing is weak writing, not AI itself
a16z crypto argues that the familiar “AI writing” feel often blamed on large language models did not start with AI at all. In a recent article, the firm said many of the traits people associate with machine-generated copy — vague but lofty phrasing, corporate-style filler, rigid structure, repetitive sentence patterns, and a missing personal voice — were already common in human writing long before generative AI went mainstream. In that view, models are not inventing a new problem so much as reproducing and scaling the most average, safest, and most predictable habits found in existing text.
The piece suggests a different way to think about AI detection. Instead of treating those traits as proof that a text was written by AI, editors and readers could use them as a checklist for bad writing. If a passage relies on fuzzy adjectives, add real people, numbers, events, and examples. If a paragraph sounds complete but says little, ask what the author is actually trying to argue. If every section reads the same, change the rhythm. The core test, a16z crypto said, is whether the piece is clear, credible, and useful.
The article also says AI can still be helpful in the writing process, especially for spotting empty passages, cutting weak sentences, and checking structure. What should not be fully handed to a model, it argues, is the writer’s own judgment about why a subject matters and which details deserve a reader’s time.
