Prompt coding is more expensive than programming ?

I have been writing code with Cursor and claude code since last 8 months,
It has been a smooth ride for projects built through prompt engineering with well documented prd and agile development docs ,But for pre-existing codebases ,particularly the frontend applications even when they already written which are documented at an architectural level, it becomes too iterative and expensive for claude code and cursor to refactor the code with incoming problems of claude code or cursor stuck in the circular loop because of no context of inter-dependencies of modules .

Comments

[Harjot Singh]

The provocative framing holds up if you only count the obvious case: throwing every task at a premium model and iterating by re-prompting is genuinely more expensive than just writing the code, especially for things you already know how to do. Prompting has a cost-per-attempt that typing doesn't.

But it's not inherent to prompt coding - it's a routing failure. Hand-coding wins when you know exactly what to write; prompting wins on the stuff you'd otherwise spend an hour googling. The expensive version is using the biggest model for everything and re-rolling. Match the tool to the task (and the model to the difficulty) and prompt coding gets cheaper than programming for the right slice of work, more expensive for the rest. The honest answer is "it depends what you're coding." Good question to raise.