What the AI engineer job market actually asks for
Eleven postings, counted. The gap is rarely the AI part.
AI engineering roles are assumed to require deep machine learning. Counting what the postings actually ask for tells a different story.
What we counted
Across eleven AI engineering postings, transcribed in full: Docker appeared in eight, Python in seven, shipping large language models to production in seven, Kubernetes in six. Model training from scratch appeared in almost none.
The job is mostly software engineering applied to a new kind of dependency. The hard part is running the thing reliably and affordably, not inventing it.
What that means if you are moving into it
If you can already build and operate backend services, you are closer than you think. If you have read a great deal about transformers but have never containerised anything, you are further away than you think, and the gap is the ordinary infrastructure part.
The skills that keep appearing
- Retrieval augmented generation, and specifically chunking and retrieval quality.
- Tool calling and structured outputs, including validation.
- Evaluation: measuring whether the output got better, which almost nobody has done.
- Cost and latency work, because inference is a line on a bill.
The honest differentiator
Evaluation harnesses are named in most serious postings and are the rarest real skill. If you can show a regression suite for prompts, with numbers that moved, you are in a small group.
Find roles that actually fit
Upload your CV and AgentFlow AI reads your field, level and skills from it, then matches you against thousands of live postings. The file is parsed in your browser and never uploaded.
Try it