Dear friends, what do you think an "AI engineer" does? I went through job posts for 2 weeks. "AI engineer" and "prompt engineer" aren't the same job anymore.
If you said "writes good prompts," fair enough. That used to be true. Today it means something else entirely. I went through job postings here in Metro Manila for the past two weeks. What I found...
"Prompt engineer" and "AI engineer" are already two different jobs.
Prompt engineer postings are exactly what they sound like. Writing and refining prompts, nothing more.
AI engineer postings ask for something bigger: someone who can build a system that makes its own decisions and takes its own action.
The job has already moved. Most people just haven't caught up. That's the real gap.
So what does "AI engineer" actually mean now? Four things.
Orchestration — chaining model calls into a working pipeline. Step one's output feeds directly into step two, then step three, without a person copying data between them. Most of the failures I've seen happen at the handoff.
Tool integration — wiring the model into real systems so it can act on what it decides: pull data from a database, hit a search API, update a record. This is the piece that turns an agent into something that actually gets work done.
Guardrails — the checks that catch a hallucinated tool call or a runaway loop before it reaches a user or a production system. Without them, failures show up in front of a customer instead of staying in the sandbox.
Evals — testing an agent across hundreds of cases, not five. This is what separates "it worked when I tried it" from "it works."
If you're still picturing "AI engineer" as a prompt job, you're about a year behind. The postings already know it's changed. The question is whether you do.
Which of these four does your team feel least confident in? Reply with a number 1-4. We'll factor it into next month's cohort.
-Tony