Why local AI feels slow on a Mac: the model re-reads the whole prompt every turn
A video argues that local AI on a Mac feels slow not because of the chip but because the model reprocesses the whole prompt each turn, which tokens-per-second counters never show. Per the checked facts, this prefill step, which handles the system prompt, tool schemas and history, is a common cause of slowness in local coding agents.
A video about it went from 37,454 to 58,389 views in 16 hours - about 6.7 times the typical pace of other AI-tool videos SteadyVane tracks on YouTube. It has 1,136 likes and 97 comments, and 58,389 views about two days after it was published.
Why it matters: If your local coding agent drags, the fix may be cache reuse, not a bigger machine. Per the facts checked, Ollama's MLX-based release for Apple Silicon adds cross-conversation cache reuse, and oMLX uses paged SSD caching to cut time-to-first-token.