Generating boring one off analysis scripts is the best use case for AI code writing, so can’t fault them for doing so. For serious projects I found LLMs—even cu

Generating boring one off analysis scripts is the best use case for AI code writing, so can’t fault them for doing so. For serious projects I found LLMs—even cutting edge Claude Sonnet 3.7 in agent mode—just can’t generate a large amount of code with the level of consistency and quality I expect and fixing such code is barely faster than writing it myself with AI autocomplete. But little scripts that I throw away after a single use? What used to take fifteen minutes or longer now takes thirty seconds.