AI engineer went from a curiosity title to 943 postings in our corpus inside a year, and 670 of them state pay. The blended US median lands at about 150,500 dollars. That much everyone suspected. What the reach table shows is less expected, and more useful if you are standing outside the field wondering about the door.
The doorway professions, counted
We compute, for every occupation, how much of an AI engineer's posted skill demand it already covers. Machine learning engineer leads, no surprise. Then comes the surprise in row two:
| Coming from | Coverage |
|---|---|
| Machine learning engineer | 59% |
| Sales engineer | 50% |
| Software engineer | 46% |
| Data scientist | 45% |
| Solutions architect | 45% |
| DevOps engineer | 39% |
| Research scientist | 34% |
| Product manager | 31% |
A sales job, half way to the hottest engineering title of the decade. It stops being strange when you read what AI engineer postings actually ask for. Yes, Python and model APIs. But also: explaining model behavior to non-technical stakeholders, scoping what a system should do, building demos, evaluating output quality against fuzzy requirements. That is half a sales engineer's week. The industry quietly needs people who can make AI systems legible to buyers and bosses, and it needs them as much as it needs another fine-tuning script.
What the postings ask for, in order
Across the AI engineer corpus, the recurring demands are working with large language models and their APIs, Python, retrieval systems and vector search, deployment and monitoring, evaluation methodology, and the connective skills: writing, stakeholder communication, and product sense. Degrees appear in postings less often than the folk wisdom claims. Portfolios of working systems appear constantly, in the requirements, in the nice-to-haves, in the interview descriptions.
An honest note on the salary number
Posted AI salaries run hot relative to official statistics; our reconciliation layer flags the gap at over 100 percent against the government anchor for the occupation family, the widest skew in our data. Some of that is real scarcity pricing. Some is asking-price inflation and remote-tech posting bias. Our 150,500 figure already blends toward the official anchor. Treat glossier numbers you see elsewhere accordingly.
Where this leaves you
If you are in one of the doorway professions, the gap between you and the title is smaller than the mythology says and it is made of specific, learnable things: LLM (large language model) application work, retrieval, and evaluation, stacked on skills you already use. Build two working systems you can show, learn to talk about their failure modes honestly, and you look like the postings. If you are not in a doorway profession yet, the table above is a map of intermediate steps. Nobody needs to start over. That is the whole point of measuring adjacency instead of guessing at it.
PivotHop July 2026 run: 943 AI engineer postings, 670 with stated pay; blended median shrinks posting percentiles toward the Bureau of Labor Statistics (BLS)Occupational Employment and Wage Statistics survey (OEWS) anchor for the occupation family (empirical Bayes, K=40). Reach percentages are destination-demand coverage over top-20 posting skills. Reconciliation deviations published in our salary method notes.