Mechanical Coding: Using ChatGPT as a bounded coding worker

The article proposes "mechanical coding," a workflow where the terminal output, rather than the AI model, serves as the source of truth for project state. This approach prevents the common failure mode where models invent context or edit incorrect files by forcing them to proceed only on verified evidence.

• The model proposes one small bounded step while the human runs it in the terminal.
• Terminal output proves whether a step passed or failed, preventing false assumptions.
• A provided template guides ChatGPT to act as a coding coordinator that performs a read-only audit first.
• The workflow requires pasting complete terminal output back to the model for validation before continuing.

This method is slower initially but saves time later by ensuring the project does not drift into invented state.