Peter Steinberger says, in the future, don't rely on manually crafting prompts for coding agents; instead, design a loop that allows the agent to prompt itself. Those still struggling with prompt words are superficially improving efficiency but are actually turning themselves into human schedulers. The risk isn't that they can't write prompts, but that they lack verification, retry, and stop mechanisms. In the end, what goes live isn't the agent, but you, the cron job that keeps retrying at 3 a.m.