AI engineer vs ML engineer: building on models, or building the models
An AI engineer builds product features on top of models trained by someone else, usually large language models reached through an API, while a machine learning engineer trains, evaluates and serves the models themselves; products built on a hosted foundation model need AI engineering first, and products that need a custom model need ML engineering.
Checked
Latent Space, the newsletter that named the AI engineer role in June 2023, drew the line between the two at the model API. On one side, ML engineers and researchers handle pretraining-scale data, training and general benchmarks. On the other, AI engineers build products on foundation models and own the product-specific data and evals. Latent Space called the line permeable, and it still is.
For a company hiring, the question is which side of the line the product's hard problem sits on. The answer decides the hire, the interview loop and, on current pay data, a large part of the budget.
What each one does
AI engineer
Wires hosted models into a product: prompts and context, retrieval over company data, tool calls and agents, evals, and the cost and latency trade-offs of each request.
ML engineer
Builds the model itself: data pipelines and labelling, training and fine-tuning, held-out evaluation, then serving, monitoring for drift and retraining.
Where the line blurs
Both roles write evals, and both care about what a model costs to run. An AI engineer who fine-tunes a model, or hosts an open-weight one, is doing ML engineering. An ML engineer who wraps a hosted model in retrieval and tools is doing AI engineering. Job posts often ask for some of both.
Interview loops show the difference in emphasis. techinterview.org describes AI engineer loops built around retrieval, a take-home with an eval set, live debugging of an agent, system design around a model the company does not control, and a customer-conversation round. ML engineer loops lean on modelling, training and data work instead.
What each one earns
On Levels.fyi, the Machine Learning Engineer title reports a US median total compensation of $282,950 (1,065 salaries, updated 31 October 2026), and the AI Engineer title $156,000 (339 salaries, updated 1 November 2026). Software engineers who tag an ML or AI focus, which includes many AI engineers at large tech companies, report $250,000 (5,921 salaries).
The figures are self-reported and the employer mix behind each title differs, so the gap says more about who holds each title today than about what the same person would be paid under either one.
Which to hire first
AI engineer first
When the product calls a hosted model: assistants, agents, document extraction, search over company data. The model already exists; the work is everything around it.
ML engineer first
When no hosted model does the job: computer vision on a production line, forecasting, recommendations, or anything trained on data that cannot leave the company.
Both, in sequence
Many products start on a hosted model and move parts of it to a fine-tuned or self-hosted model once volume, cost or accuracy justifies it.
How Motiontic approaches the choice
Motiontic's engineers work on both sides of the line. Whether a problem needs a trained model, a fine-tuned one or a hosted model with good retrieval is decided by testing on the client's own data for accuracy, latency and cost, not by which kind of engineer happens to be available.
An AI engineer compared with a machine learning engineer (checked 1 November 2026)
| AI engineer | ML engineer | |
|---|---|---|
| Starts from | A model someone else trained | Data |
| Produces | Product features and agents on that model | A trained model and its serving stack |
| Core skills | Prompts and context, retrieval, tool calls, evals | Data pipelines, training, evaluation, serving |
| Measures quality with | Product-specific evals and traces | Held-out test sets and model metrics |
| Main running cost | Model API calls per request | Training and inference compute |
| Typical failure | Plausible wrong answers, runaway agent loops | Drift between training data and production inputs |
| Levels.fyi US median total pay | $156,000 (AI Engineer title) | $282,950 (Machine Learning Engineer title) |
| Hire first when | The product runs on a hosted model | No hosted model does the job |
Common questions
What is the difference between an AI engineer and an ML engineer?
An AI engineer builds product features on models trained by someone else, usually through an API. An ML engineer trains, evaluates and serves the models themselves. Latent Space draws the line at the model API and calls it permeable.
Who earns more, an AI engineer or an ML engineer?
On Levels.fyi, the Machine Learning Engineer title reports a US median of $282,950 and the AI Engineer title $156,000, as of 31 October and 1 November 2026. Software engineers with an ML or AI focus report $250,000, so the employer mix explains much of the gap.
Can an ML engineer do AI engineering?
Often, yes. The tools overlap, though AI engineering adds product work, such as retrieval, agent design and evals for open-ended outputs, that model-training roles do not always cover.
Should a startup hire an AI engineer or an ML engineer first?
If the product runs on a hosted foundation model, an AI engineer. If it needs a model trained on its own data, such as computer vision or forecasting, an ML engineer.
Related
AI Engineer
Builds the model-facing parts of a product: prompts, retrieval, tools, evals and the code around them.
Machine Learning Engineer
Trains, evaluates and serves the models a product depends on, from the data pipeline to production inference.
AI engineer salary
What AI engineers earn by source and percentile, what moves the number, and what a hire costs.
RAG vs Fine-Tuning
Fetching the right facts at query time, compared with retraining the model's behaviour.
AI product development
New products and major AI features, built from scratch alongside the client's team.
Running agents that have to hold up?
Describe the agent, the systems it touches and what it must never get wrong. The first call is with a lab engineer, who will say plainly whether it is ready for production, and whether Motiontic is the right team.