Users end up spending significant time “arguing” with the AI model refining prompts, correcting outputs, or providing more context, only to realise manual completion would have been faster.