Total beginner here—Why is LM Studio making me do the "heavy lifting" manually?

A viral Reddit post from a self-described beginner has sparked discussion about the practical limits of running large language models (LLMs) locally. The user, Ofer1984, detailed their experience with LM Studio, a popular GUI for running open-source models like Qwen2.5-VL-7B. They prompted the model to create a simple web app and provide an easy localhost link, explicitly asking it to avoid complex developer workflows. Instead of autonomously building the project, the AI responded with a list of manual steps: placing files in specific directories, editing code, and moving assets.

This experience underscores a fundamental divide in today's AI landscape. While cloud-based agents from companies like OpenAI and Google can sometimes execute code in sandboxed environments, most locally-run models via tools like LM Studio, Ollama, or GPT4All are strictly text-completion engines. They lack the "agency" to interact with a user's operating system, file system, or network ports due to massive security and complexity constraints. The post serves as a real-world case study in managing expectations: local LLMs are powerful for conversation and code generation, but executing that code remains a manual, human-led process.