Have you ever pleaded with a large language model to follow its prompt more precisely by asking it to be more helpful, repeating yourself, SHOUTING AT IT, manipulating or threatening it? Can we call this “software engineering” and what can we guarantee about the resulting systems? Large language models enable software to be built in powerful new ways, but their use turns many notions of software engineering on their head. We explore how to maintain an engineering mindset whilst taking advantage of new opportunities with language models, based on a decade of experience researching and developing LLMs and incorporating them into shipping products.