The short answer
An LLM engineer builds product features on top of existing foundation models: retrieval, prompting, tool use, agents and the evaluations that keep them reliable. An ML engineer trains, tunes and serves models on your own data. "AI engineer" is an umbrella title that, in 2026, usually means the LLM-engineer profile. Hire for the work: building with models, or building the models themselves.
Key takeaways
- An LLM engineer (often titled applied AI engineer or AI engineer) builds with models. An ML engineer builds the models.
- "AI engineer" is not a fixed job. Some companies use it for LLM work, others for classic machine learning, so the brief matters more than the title.
- The skills overlap in software engineering and evaluation, but the daily work, the data and the interview loop are different.
- An MLOps or AI platform engineer is a third role: they build the infrastructure both of the others ship on.
- For a first AI hire that adds AI features to an existing product, a strong LLM engineer with solid general engineering is usually the right start.
- Post the title that matches the work, and describe the work in the first lines of the job ad. Most mis-hires start with the wrong title.
On this page9 sections
Why the titles got confusing
The title "AI engineer" grew faster than any shared definition of it. LinkedIn's Jobs on the Rise 2026 ranked AI engineer as the fastest-growing job in the United States, based on jobs members started between January 2023 and July 2025. The skills LinkedIn associates with it, such as LangChain, retrieval-augmented generation (RAG) and PyTorch, mix LLM application work with classic deep learning, and people moved into the role mostly from software engineering, data science and full-stack roles [1].
The modern meaning comes from 2023. An essay on Latent Space, The Rise of the AI Engineer, argued that a new kind of engineer was building products on top of foundation models through APIs, and that this was a different job from machine learning engineering. It quotes Andrej Karpathy: "One can be quite successful in this role without ever training anything" [2].
So today the same title can describe two people with very different skills. That's why this is one of the first things we settle with a company before a search starts. For the full hiring process, see our complete guide to hiring AI engineers.
Definitions
An LLM engineer is a software engineer who builds product features on top of large language models and other foundation models. They design retrieval, prompts and context, tool calling and agent logic, then build the evaluations and guardrails that keep those features dependable for real users. They usually work in the product codebase and rarely train models. Common titles: LLM engineer, applied AI engineer, AI engineer, generative AI engineer. See what an LLM engineer does.
A machine learning (ML) engineer is an engineer who trains, tunes and serves models. They turn a modelling problem into data, features, training runs and an evaluation plan, then put the model behind a reliable service and keep it accurate as the data changes. Their work includes fine-tuning pretrained and foundation models when a task needs it. See what a machine learning engineer does.
"AI engineer" is an umbrella title. In 2026 it most often means the LLM-engineer profile above, but some companies still use it for ML engineers, data scientists or research engineers. Treat it as a label to clarify, not a definition.
LLM engineer vs ML engineer: side by side
The clearest way to tell them apart is to look at what each one changes when something goes wrong. An LLM engineer changes the retrieval, the prompt, the tools or the evaluation set. An ML engineer changes the data, the features, the model or the training run.
| Role | Main job | Core skills | Typical deliverables | How they measure quality | Where they sit |
|---|---|---|---|---|---|
| LLM engineer (applied AI engineer) | Build product features with existing models | RAG, embeddings and search, prompt and context design, tool calling, agents, evaluations, guardrails, Python or TypeScript | Assistants, search, extraction, agents, AI features in the product | Task-specific evaluation sets, model-graded and human review, regression checks on every change | Product engineering team |
| Machine learning engineer | Train, tune and serve models | PyTorch or TensorFlow, data preparation, feature engineering, training and fine-tuning, model serving | Ranking, recommendation, forecasting, classification and fine-tuned models | Offline metrics chosen for the business problem, error analysis, online tests | ML or data team, sometimes embedded in product |
| MLOps or AI platform engineer | Build the platform models ship and run on | Pipelines, model registry, Kubernetes, CI/CD for models, GPU capacity, monitoring, cost control | Training and deployment pipelines, serving infrastructure, observability | Reliability, deployment speed, cost per model or request | Platform or infrastructure team |
Two things the table doesn't show. First, both LLM and ML engineers need strong general software engineering: production code, tests, code review. Second, both care about evaluation. The difference is what they evaluate. An LLM engineer mostly evaluates a system's outputs on real tasks. An ML engineer mostly evaluates a model against a metric on held-out data.
Which one do you need?
Start from the problem, not the title. Four common scenarios:
| Your project | Hire first | Why |
|---|---|---|
| A support assistant or internal search over your documents (RAG) | LLM engineer | The work is retrieval, prompts, evaluation and guardrails on top of an existing model |
| An agent that takes actions in your product (booking, filing, triage) | LLM engineer, with strong backend skills | Tool design, permissions and failure handling matter more than model training |
| A ranking, recommendation or fraud model trained on your own data | ML engineer | You need data, features, training runs and offline evaluation |
| Fine-tuning a model because prompting and retrieval aren't enough | ML engineer (or an LLM engineer with fine-tuning experience) | Data curation and training are the core of the work |
| Several teams shipping models, slowed down by deployments and GPU capacity | MLOps or AI platform engineer | The bottleneck is infrastructure, not models |
| Clean, governed data for analytics, ML and AI features | Data engineer | Every model is limited by the data it gets |
If you're hiring your first AI person to add AI features to an existing product, the LLM engineer is usually the right first hire. They can ship something useful on existing models while you learn where a specialist would help.
Where MLOps fits
MLOps is the third role people mix up with the other two. Google Cloud describes MLOps as an engineering practice that unifies ML system development and ML system operation, and notes that only a small fraction of a real-world ML system is the ML code itself [3]. A widely cited NeurIPS paper made the same point in 2015: the surrounding infrastructure (data collection, configuration, serving, monitoring) is where most of the complexity and technical debt sits [5].
An MLOps or AI platform engineer owns that surrounding system: pipelines, deployment and rollback, GPU capacity, monitoring and cost. You need one when more than one team ships models, or when getting a model into production takes longer than building it. See what an MLOps engineer does and our guide on how to hire MLOps engineers.
Overlaps and hybrids
The boundaries are real but not absolute. Common hybrids:
- The ML engineer who moved into LLM work. Strong on evaluation and data, sometimes less used to product engineering pace. Often excellent for fine-tuning and evaluation-heavy LLM work.
- The full-stack engineer who became an LLM engineer. Strong on product, APIs and shipping. Check that they measure quality systematically, not by eye.
- The research engineer. Comfortable reading papers and training models, and a good fit when you are pushing a model further than off-the-shelf use. See the AI researcher profile for the research end of the spectrum.
One person can cover both roles in a small team, especially early on. The risk is asking one person to do both at production depth. Training pipelines and LLM product features each take sustained attention once real users rely on them.
How to evaluate each role
The interview loops differ because the evidence differs. AI coding tools are standard now: in Stack Overflow's 2025 survey, 84% of developers said they use or plan to use AI tools, while more of them distrust the accuracy of those tools (46%) than trust it (33%) [4]. So for both roles, test judgement rather than recall.
- LLM engineer: ask about an LLM feature real users relied on, how they measured quality beyond "it looked right", a production failure they caught, and the trade-offs between quality, latency and cost.
- ML engineer: ask about a model they took from experiment to production, the metric they chose and why, and error analysis that changed the data or the model rather than just the hyperparameters.
- MLOps engineer: ask about a platform other engineers actually adopted, a deployment path with rollback, and a cost or capacity problem they diagnosed.
Our interview framework for AI engineers turns these into a full loop with a scoring rubric.
Title confusion in job ads
Three habits attract the wrong candidates:
- A generic "AI engineer" title with an ML job underneath. LLM application engineers apply, then fail an interview about training pipelines. Or the reverse.
- A list of every framework in the field. It signals that the team hasn't decided what the person will build. Name the problem and the production constraints instead.
- "Years of LLM experience" requirements. Modern LLM application work is only a few years old. Ask for shipped work instead, such as "built and maintained an evaluation suite for an LLM feature".
Our LLM engineer job description template shows how to write the ad around the work.
How Tailored Minds helps
The six AI role profiles on our site are the working definitions we use with companies. When you brief a role, we help you decide which one you actually need before the search starts. Then our team searches the talent network while approved independent recruiters can work the role at the same time. People review every candidate we shortlist against the signals for that specific role, so an LLM engineer is judged on LLM work and an ML engineer on ML work. AI assists the search; people make the decisions. Hire AI engineers with Tailored Minds.
Frequently asked questions
Do LLM engineers need to train models?
Is an AI engineer the same as an ML engineer?
Is prompt engineering a separate job?
What's the difference between MLOps and ML engineering?
Can one person be both an LLM engineer and an ML engineer?
Sources
- LinkedIn Jobs on the Rise 2026: the fastest-growing jobs in the U.S. (reported by CNBC), CNBC
- The Rise of the AI Engineer, Latent Space
- MLOps: Continuous delivery and automation pipelines in machine learning, Google Cloud
- 2025 Developer Survey: AI, Stack Overflow
- Hidden Technical Debt in Machine Learning Systems (Sculley et al., 2015), NeurIPS Proceedings
About the author

Tercio Lima
Growth Lead at Tailored Minds
Tercio Lima is the Growth Lead at Tailored Minds. He runs the brand, content and search work that brings companies to the platform, and the sourcing work that finds engineers for their roles, so what he writes about hiring comes from live searches rather than desk research.
Share this article
