It's a real skill and a fake job title, mostly. Both things at once:
Real: knowing how to get reliable outputs from LLMs — clear context, examples, output formats, breaking tasks into steps, evaluating failure modes — measurably matters when you're building products on top of these models. Developers who can do this ship better AI features.
Fake-ish: 'prompt engineer' as a standalone career is already dissolving into other roles, the way 'webmaster' did. It's becoming a skill every developer, marketer and analyst is expected to have, not a job unto itself. And models keep needing less prompt gymnastics with every generation.
Worth learning? Yes — in a weekend from free resources (Anthropic's and OpenAI's own prompting guides are excellent). Worth $500 for a certificate? Absolutely not. Nobody hiring has ever asked for one.