AI is antithetical to learning

Spend enough time with LLMs and you start noticing certain patterns. Maybe you’re working on a codebase and you want to upgrade a dependency or language version, redesign some component, or add new features. You ask the LLM how to do it, and it gives you a brief description and then finishes with “Do you want me to start?”. Sometimes it doesn’t even bother responding, it just edits the code for you. You ask it for the commands to look up DNS records for a domain, and it responds with the DNS records. Not the commands. And it always finishes by suggesting another thing it could do. Guiding you forward, but always keeping you one step removed from what is actually going on. If you’re very disciplined you can stay involved, by carefully crafting your prompts, inspecting all the actions taken in the harness (actions that are less visible with every harness update), and by stopping the agent when it goes off the rails. But it requires just that, discipline. All that willpower to stay…