What is the difference between an AI engineer and an ML engineer?

The ML engineer job grew up around supervised learning. You collect data, define labels, train a model, evaluate it on a held-out set, and deploy it to a service that returns predictions. The hard problems are data quality, distribution shift, and serving latency. ML engineers usually work next to data scientists and data engineers, and the system runs models the team owns end to end. The discipline is roughly 15 years old at industrial scale and has well-established tooling: scikit-learn, PyTorch, TensorFlow, MLflow, Weights and Biases, and the major cloud ML platforms.

AI engineering, in the modern sense, grew up around large language models that someone else trained. Your raw material is a pretrained foundation model accessed by API or running on your own GPUs. The hard problems are prompt design, retrieval quality, agent reliability, evaluation against real user goals, and cost per request. AI engineers usually work next to product managers and security engineers, because the system touches users directly and the failure modes are public. The discipline emerged in roughly 2022 after GPT-3.5 and Claude reached production-quality and is still maturing.

The skill stacks share a foundation but diverge in the middle. Both roles expect comfort with Python, version control, and production deployment. ML engineers go deeper into PyTorch or TensorFlow, training loops, distributed training (DeepSpeed, FSDP), the ML ops stack (feature stores, model registries, drift monitoring), and the math-heavy debugging of models that learned the wrong thing. AI engineers go deeper into the API surface of frontier models, retrieval (vector databases like Pinecone, Weaviate, Qdrant; embedding models), agent frameworks (LangChain, LlamaIndex, AutoGen), evaluation methodology (HELM, MMLU, custom benchmarks), and the latency-cost tradeoffs of inference.

Compensation overlaps but is not identical. Per Levels.fyi April 2026 bands, ML Engineer total compensation at large US tech employers runs $230,000 to $420,000 for senior individual contributors. AI Engineer roles in the same firms pay slightly more on average ($250,000 to $450,000 senior IC band), partly because demand is hotter and partly because the role is newer and harder to staff. Frontier labs (OpenAI, Anthropic, Google DeepMind) pay above either band, especially for research scientist roles where total compensation routinely exceeds $700,000 and reaches into seven figures.

Hiring signals differ. ML engineers get screened on linear algebra, probability, ability to debug a model that learned the wrong thing, and standard ML interview topics (gradient descent variants, regularization, bias-variance tradeoffs, evaluation metrics for different problem types). AI engineers get screened on system design around language models, eval design, the ability to reason about failure under prompt injection or hallucination (per OWASP LLM Top 10), and increasingly the security and safety implications of agentic systems. Both interviews include coding, but the conceptual depth is in different places.

Educational backgrounds skew differently. Per the 2024 Kaggle State of Data Science and ML survey, 73 percent of working ML engineers hold a master's degree or PhD, with strong concentrations in computer science, statistics, applied math, and operations research. AI engineer educational backgrounds skew toward bachelor's-level computer science plus self-directed AI study. Both paths produce strong practitioners; the difference is structural rather than capability-based.

Some teams use the titles interchangeably and pay no attention to the distinction. Read the job description before assuming. If the role talks about retrieval-augmented generation, agents, and prompt evaluation, it is AI engineering even if the title says ML. If it talks about training pipelines, feature engineering, and model registries, it is ML engineering even if the title says AI. The MLOps versus AIOps split increasingly mirrors this distinction, with MLOps focusing on training-and-deployment for custom models and AIOps focusing on inference-cost and serving-throughput for foundation models.

If you are choosing a track, the AI engineering market is hiring faster right now, the work compounds with software engineering experience, and the compensation premium is real. If you have a research or applied math background and enjoy training rather than composing systems, ML engineering may be the better fit. The cybersecurity convergence rewards either path: AI Red Team Engineer, AI Safety Engineer, and AI Security Engineer roles draw from both stacks, with AI engineering background mapping more directly to security work and ML engineering background mapping more directly to evaluation and alignment research.

Honest tradeoffs. ML engineering is more durable across hype cycles because supervised learning use cases (recommendations, fraud detection, ranking, demand forecasting) continue regardless of LLM trends. AI engineering is hotter right now but more exposed to platform consolidation: if the major foundation-model providers move into more layers of the application stack, some AI engineering work compresses. Both paths reward T-shaped expertise: deep in one area, working knowledge of the other. DecipherU's Applied AI career guides cover both tracks with sub-discipline detail.