How I use AI (Oct 2025) - Ben Stolovitz

Ben Stolovitz takes stock of how he uses AI and says it is most transformative in coding, especially autocomplete. He also uses LLMs for research, summaries, transcription, and editing, while staying skeptical about their reliability in search, writing from scratch, and image generation.

Reading notes

#

• AI has changed coding the most for him, mainly through autocomplete that predicts the next line and helps with familiar patterns and syntax.
• He finds agent mode useful for some common tasks, but agentic pull requests are slowed down by validation and frequent errors.
• AI code review can catch real bugs, but it also produces noisy or unhelpful comments.
• For research and search, he values LLMs most for “pub facts,” finding hard-to-remember sources, and explaining well-known concepts.
• He distrusts LLMs for traditional search, product search, and literature search because they can hallucinate, agree too easily with premises, and mislead users with citations.
• He considers summarization and transcription very strong uses of LLMs, especially for reducing long meetings to short summaries.
• He does not use AI to write from scratch and prefers to keep his own voice in work writing and blogging.
• He finds AI editing surprisingly effective, but sometimes avoids its exact phrasings because they can feel too strong or too close to replacing his voice.
• He has not found image generation very useful and dislikes AI art when he notices it.
• He thinks music generation can be enjoyable and creative, but in practice may be used for laziness, stinginess, and grift.
• He expects AI to become normal and pedestrian over time, with a mix of useful, mediocre, bad, and scary effects.