AI Engineer Job Description 2026: Skills, Salary, and What the Role Actually Is

Between 2023 and 2026, "AI engineer" became one of the fastest-growing job titles in tech. It also became one of the least precise. Companies use it to mean everything from an engineer who calls an OpenAI API to someone building custom transformer architectures from scratch. The result is job descriptions that are a confused mix of requirements from multiple distinct roles — and pipelines full of candidates who match some requirements but not the actual job.

Getting this right matters because the three types of AI engineer are genuinely different roles: different skills, different vetting processes, different compensation, and different profiles of person who will succeed. Here is how to tell them apart and write a job description that attracts the right one.

The Three Types of AI Engineer in 2026

Type 1: AI Application Engineer

This is the most common hire in 2026. AI Application Engineers build products on top of existing foundation models — LLM APIs, multimodal models, embedding models — without training custom models from scratch. Their work includes:

• Designing and implementing RAG pipelines and retrieval systems
• Building LLM-powered features: chat, summarisation, classification, extraction
• Prompt engineering and evaluation at scale
• Integrating with vector databases (Pinecone, Weaviate, pgvector)
• Agent and tool-use architecture
• Evaluation frameworks for LLM output quality

Core skills: Python, LLM APIs (OpenAI, Anthropic, Gemini), LangChain or similar orchestration, vector databases, REST API design, basic MLOps for serving.
Salary (US, senior): $150K–$230K base

Type 2: ML Engineer

ML Engineers train, fine-tune, and deploy custom machine learning models. They work closer to the model layer than Application Engineers, and closer to production than Research Engineers. Their work includes:

• Training and fine-tuning models on custom datasets
• Building and maintaining data pipelines for ML
• Model serving, monitoring, and reliability in production
• Distributed training and GPU infrastructure
• Feature engineering and experiment tracking
• MLOps: CI/CD for models, A/B testing, model versioning

Core skills: PyTorch or JAX, distributed training (FSDP, DeepSpeed), model serving (TorchServe, TGI, vLLM), MLflow or similar, Kubernetes, strong Python and software engineering.
Salary (US, senior): $180K–$280K base

Type 3: Research Engineer

Research Engineers sit at the boundary between ML research and production engineering. They implement novel architectures from papers, run large-scale experiments, and translate research ideas into production-ready systems. Their work includes:

• Implementing and evaluating novel model architectures
• Running large-scale training experiments on hundreds of GPUs
• Bridging the gap between research papers and production code
• Ablation studies and empirical analysis of model behaviour
• Contributing to internal research publications

Core skills: Deep ML theory, JAX or PyTorch, CUDA-level understanding of GPU computation, strong mathematical foundations, publication record or equivalent depth of research engagement.
Salary (US, senior): $220K–$350K+ base

The Most Common Job Description Mistakes

Mixing requirements from all three types

The most frequent mistake is writing a job description that lists requirements from all three role types simultaneously: "experience with PyTorch, fine-tuning LLMs, building RAG pipelines, CUDA optimisation, and publishing ML research." This person does not exist — or if they do, they are extremely rare and will not be interested in most companies. The result is a pipeline of candidates who partially match and a role that cannot be filled.

Requiring a PhD for non-research roles

A PhD is genuinely necessary for Research Engineers whose work involves novel architecture development. It is not necessary for ML Engineers or AI Application Engineers — and requiring it eliminates a large pool of strong practitioners who have built substantial production experience without a research background.

Listing technology buzzwords without specifics

"Experience with AI/ML technologies" tells strong candidates nothing. Specific requirements — "experience fine-tuning Llama-class models on domain-specific datasets with LoRA or QLoRA" — signal that the company knows what it is doing and attract candidates who can actually do it.

Vague seniority signals

"5+ years of experience in machine learning" as the only seniority signal is too blunt. Strong job descriptions specify what senior looks like in practice: "has taken at least two models from experiment to production serving 1M+ requests/day" or "has led the ML architecture decisions for a team of 3–5 engineers."

Job Description Templates by Role Type

AI Application Engineer

What you will do:

• Design and build LLM-powered features across our product (summarisation, extraction, chat)
• Architect and maintain our RAG pipeline and retrieval infrastructure
• Build evaluation frameworks to measure and improve LLM output quality
• Own the integration between our product and foundation model providers

What we are looking for:

• Strong Python engineering — you write clean, tested, maintainable code
• Hands-on experience with LLM APIs and prompt engineering in production
• Experience building RAG systems with vector databases
• Understanding of LLM evaluation — you can design metrics, not just eyeball outputs
• Track record of shipping LLM features to users, not just prototyping them

ML Engineer

What you will do:

• Train and fine-tune models on our proprietary datasets
• Build and maintain ML infrastructure: training pipelines, feature stores, model serving
• Own model performance in production — monitoring, alerts, retraining triggers
• Work closely with data scientists to take prototype models to production

What we are looking for:

• Strong PyTorch experience — you can write custom training loops, not just use high-level wrappers
• Experience with distributed training and GPU infrastructure
• Production ML engineering experience — models you have shipped and maintained at scale
• MLOps: experiment tracking, model versioning, CI/CD for model updates
• Solid software engineering fundamentals — testing, code review, system design

Interview Framework by Role Type

The interview process should match the role type:

• AI Application Engineer: System design for an LLM-powered feature + LLM evaluation exercise + code review of a RAG implementation
• ML Engineer: Training pipeline design + production debugging scenario + model serving architecture + code quality assessment
• Research Engineer: Deep paper discussion + novel problem formulation + architecture critique + implementation of a component from a recent paper

For a full technical vetting process, see our ML engineer technical hiring framework and the LLM engineer vs ML engineer comparison.

Salary Benchmarks (2026, US market)

• AI Application Engineer (Senior): $150K–$230K base; $200K–$320K TC at well-funded startups
• ML Engineer (Senior): $180K–$280K base; $250K–$400K TC
• Research Engineer (Senior): $220K–$350K base; $300K–$500K+ TC at AI labs

UK equivalents: Application Engineer £80K–£130K; ML Engineer £100K–£160K; Research Engineer £130K–£200K.

Working with VAMI on Job Description and Search

VAMI writes the job description together with you before starting the search — diagnosing which of the three AI engineer types fits your stage, architecture, and team composition. This role clarity step cuts time-to-hire by eliminating mismatched applicants from day one.

If you are unsure which type you need, or if your current JD is not generating the right pipeline, start with a scoping call.

Summary

• "AI engineer" in 2026 covers three distinct roles: Application Engineer, ML Engineer, Research Engineer
• Each requires different skills, earns different compensation, and needs a different interview process
• The most common mistake is mixing requirements from all three types in one JD
• PhD requirements only belong in Research Engineer roles
• Salary ranges: $150K–$230K (Application), $180K–$280K (ML), $220K–$350K+ (Research) in the US
• Write the JD around what the person will actually do in week one — not a wishlist of credentials