Retyping LLM Code by Hand to Fight Cognitive Debt

Copy-pasting AI code makes you faster today, but it leaves you clueless tomorrow.

Developer Ankur Sethi shared a low-tech trick to stop AI brain fog. He forces his coding assistant to output proposed changes in chat, then retypes every line into his editor by hand. Sethi configures agent instruction files to explicitly ban the AI from creating, editing, or running commands on project files. Instead of a 10x speed boost, he gets a 2x boost—and keeps full mental control over his codebase.

Why it matters: Letting AI one-shot entire features builds up massive cognitive debt. When code lands directly in your editor, you lose your spatial map of the system and blind-review messy PRs. Manual typing forces you to slow down, spot hallucinations early, and clean up bad design before it hits git.

Try it: Add rules like these to your AI agent files:

• Never edit, create, or delete files directly. Show proposed edits in chat so I can type them in manually.
• Show terminal commands in chat instead of running them.
• Do not explain basic syntax or concepts unless explicitly asked.

It turns out the classic advice for learning to code—never copy-paste from the tutorial—is still the best strategy for staying sane around LLMs.