The phrase 'AI Engineer' now appears on roughly one in twelve engineering job listings. It means almost nothing by itself — and almost everything once you dig into the actual requirements.
The ten archetypes we found
After tagging 4,200 listings by required skills, reported comp, and team structure, ten distinct role shapes emerged — from model-infrastructure engineers building inference clusters, to application engineers gluing APIs together, to research engineers sitting one step below ML researchers.
Tip
The single best signal: look at the tech stack listed under 'nice to have.' It reveals the team's actual ambitions better than the job title does.
What they actually pay
Median comp for the infrastructure-heavy roles sits 18-24% above application-layer roles at the same level, largely driven by the scarcity of engineers who can work across CUDA, serving infra, and distributed training simultaneously.
The ceiling is genuinely high. The top decile of ML infrastructure roles at large labs reported total comp above $600k, and those numbers have not compressed meaningfully despite broader tech layoffs.