LLM code dependence triggers a manual retyping backlash

A thread on manually retyping LLM-generated code to prevent 'cognitive debt' got traction, and the responses ranged from 'this is prayer, not engineering' to genuine agreement that copy-pasting code you don't understand is compounding a skill deficit. The thread on 'meat proxy' thinking connected directly: if you use LLMs to offload all judgment, you get worse at judgment.

This connects to a broader cluster of threads today. The note-taking and PKM discussion surfaced the same anxiety: people are building elaborate systems, or now delegating to AI, to avoid confronting that they are not actually processing or retaining anything. The developer tool attachment thread made the same point from a different angle, that trust in tools is earned through understanding, not convenience.

The pattern across all three: there is a growing, specific anxiety among builders that AI-assisted workflows are creating hidden skill debt that will not be visible until it matters. This is not a generic AI-is-bad argument. It is a practical concern about what happens when the AI is wrong and you have lost the ability to catch it.