What an MLOps Engineer Does (and What It Pays)

Every company with a promising AI demo eventually hits the same wall. The model works on a laptop, the board loves the prototype, and then someone asks the awkward question: how do we run this for a million customers, every day, safely, without it quietly getting worse? The person who answers that question is the MLOps engineer.

I've watched this role go from an obscure DevOps sub-speciality to one of the most important jobs in AI. It's not the flashiest title, but it's where AI stops being a science project and starts making money. In this edition of our What Each Role Does series, I'll break down what MLOps engineers actually do, the tools they use, how to get in, and what the real, published pay data says.

What an MLOps engineer actually does

MLOps (machine learning operations) is the engineering discipline of getting models into production and keeping them healthy once they're there. Think of it as DevOps for models, with one big twist: software doesn't usually decay on its own, but models do. The world changes, the data drifts, and a model that was brilliant in March can be making expensive mistakes by September.

A real job ad shows the scope well. PathAI's recent Senior/Staff Software Engineer, ML Ops posting, listed on the AI Hiring Index, asked the engineer to build infrastructure and automation in AWS and on-premises for ML development and deployment, lead architecture for the company's MLOps suite with security and compliance in mind, evaluate new MLOps tools, and automate CI/CD for models, feature pipelines and deployment strategies using Kubernetes, Airflow and other orchestration tools. It also asked for experience with model lifecycle management, feature stores, model monitoring, and streaming data tools such as Kafka, Flink or Spark Streaming.

In plain English, a typical week might include:

• Building pipelines that take a data scientist's training code and turn it into a repeatable, versioned, automated process.
• Deploying models behind APIs, with rollouts, rollbacks and A/B tests, just like any other production service.
• Monitoring latency, cost, accuracy and data drift, and setting alerts for when a model starts misbehaving.
• Retraining models on fresh data, automatically where possible.
• Managing GPUs and cloud spend, which is a serious job when inference bills can rival payroll.
• Governance: tracking which model version made which decision, on which data, for audit and compliance.

That last one is growing fast. In 2026 the role increasingly covers large language models too, so you'll see titles like "LLMOps" alongside MLOps. AWS's updated machine learning engineer certification now lists ML engineer, MLOps engineer and LLMOps engineer side by side as target roles, which tells you where the industry thinks this is heading.

The skills and tools that matter

You don't need to be a research scientist. You need to be a strong engineer who understands the machine learning lifecycle. The core toolkit looks like this:

• Python and SQL, fluently. These are table stakes.
• Containers and orchestration: Docker and Kubernetes come up constantly.
• Pipeline and workflow tools: Airflow, Kubeflow and MLflow are common names, along with the managed ML platforms on each major cloud.
• Cloud platforms: AWS, Google Cloud or Azure, plus infrastructure as code.
• CI/CD for both code and models.
• Data engineering basics: feature stores, streaming and data quality checks.
• Monitoring and observability for model performance and drift.
• LLM serving and evaluation, increasingly, including retrieval pipelines and prompt and model versioning.

AI coding assistants are part of the job now too. PathAI's ad explicitly asked for experience using AI assistants in development. Tools like Claude, OpenAI's models or Grok are all reasonable to experiment with for writing pipeline code, tests and infrastructure templates. The skill that matters is knowing when the output is wrong.

The underrated skill? Communication. You sit between data scientists, platform teams, security and the business. The best MLOps engineers I know can explain to a product owner, in two sentences, why a model needs retraining and what it'll cost.

What MLOps engineers earn

Here's where I want to be careful, because salary sites measure very different things. Some use self-reported salaries, some use advertised ranges in job postings, and some include bonus and equity. So here are several real, published figures side by side, all US dollars and US-focused.

• Glassdoor: average pay of $149,506 a year, with a typical 25th to 75th percentile range of $117,827 to $191,675 and the 90th percentile at $238,251. That's based on a relatively small sample of 69 anonymous salary submissions, so treat it as a guide.
• Salary.com: an average of $130,595 as of June 2026, with a 25th to 75th percentile range of $116,608 to $139,112.
• AI Hiring Index (Made by Agents, from Adzuna listings): 152 open MLOps engineer postings with an average advertised salary of $182,590 as of 26 October 2026.
• Data Pay Index: a $205K median across 578 US postings with pay that mention MLOps (seen since 18 October 2026). By level: senior $195K, staff/lead $235K, principal $237K, director and above $244K.

What do I take from that? Survey-style sites cluster around $130K to $150K, while live job postings that mention MLOps skew higher, around $180K to $205K, partly because the roles being advertised right now are heavily senior. Data Pay Index had too few entry-level postings with pay to show a figure, which says a lot.

Individual ads tell the same story. PathAI's Senior/Staff ML Ops role advertised base pay of $128K to $196K, and its Associate Director, MLOps Engineering role on the AI Hiring Index shows $182K to $278K.

Who's hiring

Demand is broad. Data Pay Index counts 1,541 open data and AI postings at 809 companies mentioning MLOps, about 11.7% of all the data and AI postings it tracks. Most sit inside machine learning engineer roles (622 postings), followed by AI engineer and data scientist roles. So many MLOps jobs aren't titled "MLOps" at all.

On HireIndex's MLOps list this week, the hiring names include Apple, Amazon, Reddit, Databricks, Bloomberg and Atlassian, but also JPMorganChase, PNC, HCA Healthcare, Travelers, Booz Allen Hamilton and Trainline. That mix of banks, healthcare and defence matters. As I wrote in Banks and Hospitals Are Outbidding Big Tech for AI Engineers, banks and hospitals are now paying premiums for AI-skilled engineers, and MLOps is exactly the kind of skill they need to run AI safely in regulated environments. Remote is common too: HireIndex counted 73 of its 102 listed MLOps roles as remote.

How to get in

MLOps is rarely anyone's first job. Most people arrive from one of three directions:

1. DevOps, SRE or platform engineering. You already know Kubernetes, CI/CD and on-call. Learn the ML lifecycle: training, evaluation, drift, feature stores. This is the most common path I see.
2. Data engineering. You know pipelines and data quality. Add model deployment, serving and monitoring.
3. Data science or ML engineering. You know the models. Add production engineering discipline, which is often the harder half.

Whichever door you use, build one end-to-end project and put it on GitHub: ingest data, train a simple model, track experiments, deploy it behind an API in containers, add monitoring and a drift alert, and set up automated retraining. Then do it again with a small open-weight language model and an evaluation suite. That single project will do more in an interview than any list of buzzwords.

Certifications worth considering

Certifications won't replace experience, but they're a useful signal, especially if you're switching from another field. Two that map directly to MLOps work:

• AWS Certified Machine Learning Engineer – Associate. AWS is updating this exam: the last day for MLA-C01 in English was 28 September 2026, and the updated version is in beta, with MLOps and LLMOps engineers named as target roles.
• Google Cloud Professional Machine Learning Engineer. Google's exam page lists automating and orchestrating ML pipelines, serving and scaling models, and monitoring AI solutions among the skills tested. It's a two-hour exam costing $200, valid for two years, with three or more years of industry experience recommended.

A Kubernetes certification is also a sensible addition if you're coming from the data side.

My take

I'm genuinely bullish on MLOps. Every AI model that makes it into production creates ongoing work: monitoring, retraining, cost control and governance. That's not a one-off project, it's a permanent operating function, and as AI spreads into banks, hospitals and insurers, the number of production models only goes up.

It's also a role AI tools make more productive rather than redundant. Agents can write pipeline code, but someone still has to own reliability, cost and risk when real customers depend on the output. If you're a DevOps or data engineer wondering how to ride the AI wave without becoming a researcher, this is your on-ramp. Browse the latest MLOps and AI jobs and see what's open today.

Sources

• AI Hiring Index: PathAI Senior/Staff Software Engineer, ML Ops
• Glassdoor: MLOps Engineer salaries
• Salary.com: MLOps Engineer salary (June 2026)
• Made by Agents: AI Hiring Index, open roles and salaries (26 Oct 2026)
• Data Pay Index: MLOps jobs, demand and pay (October 2026)
• HireIndex: open MLOps Engineer roles (updated 26 Oct 2026)
• AWS Certified Machine Learning Engineer – Associate
• Google Cloud Professional Machine Learning Engineer