Prompt Engineering Was Never Going to Be a Standalone Job Title

For about eighteen months there were job postings for 'Prompt Engineer' with six-figure salaries attached, and I remember thinking the title had an expiration date stamped on it that nobody wanted to read. It did. Those roles either got folded into broader AI engineering positions or quietly vanished, and the people who held them either grew into something bigger or found the role didn't exist anymore.

Writing a good prompt was never the hard part

The actual difficulty in shipping an LLM feature was never finding the right phrasing — it was building the eval harness to know if a phrasing change helped, designing the retry and fallback logic for when the model does something unexpected, structuring the output so downstream code doesn't break, and managing cost and latency at scale. Prompt wording is maybe 10% of that work, and it's the 10% that's easiest to learn on the job in an afternoon.

A job title built around the easiest 10% of a skill set doesn't survive contact with the other 90%.

What actually happened to the skill

It didn't disappear — it got absorbed into "AI engineer" or "ML engineer," the same way "writing good SQL" didn't become a standalone job title even though it's genuinely difficult and genuinely valuable. It's a skill every competent person on the team needs, not a role you hire separately. I've never once seen a team succeed with a dedicated prompt engineer sitting apart from the engineers who understood the system the prompt lived inside.

If you're early in your career and were chasing that title specifically, don't panic — chase the broader skill set instead. Learn to build the eval harness, the retry logic, the structured-output parsing, the cost monitoring. The prompt wording will come along for free, and you'll have a role that still exists in two years.