What a Machine Learning Engineer Does (and Pays)

11 November 20269 min readBrett Iredale

What a machine learning engineer does day to day, how it differs from MLOps and data roles, the skills and certifications that matter, and real 2026 pay data.

If I had to pick one job title at the centre of the AI boom, it would be machine learning engineer. It's the role that takes a model from "interesting notebook" to "feature millions of people use". The fraud check on your card payment, the recommendations in your shopping app, the perception system in a self-driving car: an ML engineer built and shipped those.

It's also one of the most misunderstood titles, because it overlaps with so many neighbours. So I'll be precise about what an ML engineer does, how it differs from the MLOps, data and inference engineers we've covered in this series, the skills and certifications that matter, what it really pays, and how to break in.

What a machine learning engineer actually does

In one line: a machine learning engineer designs, trains, evaluates and ships the models that solve a specific business problem, and owns how well those models perform once they're live.

That word "owns" matters. Does the fraud model catch more fraud without blocking good customers? Did the new version beat the old one on the metrics that matter? That's the ML engineer's problem to solve.

A typical week looks something like this:

Framing the problem. Turning "we lose too much to fraud" into a measurable ML task with a label, a target metric and a baseline to beat.
Building features and datasets. Choosing the signals the model sees and creating clean training, validation and test splits that don't leak the answer.
Training and tuning models. Running experiments in PyTorch, TensorFlow or scikit-learn, and fine-tuning pre-trained models where that's faster.
Evaluating honestly. Offline metrics, then A/B tests, plus checks for bias, drift and edge cases.
Shipping to production. Packaging the model and working with platform and MLOps colleagues so it runs reliably at scale, then retraining when the world changes.

A real example: Plaid announced a new Seattle engineering office this week, and said the largest group of its open Seattle roles is machine learning and data, including senior ML engineers on fraud and embedded insights (Plaid). A hard business problem, lots of data, and a model that has to keep getting better.

ML engineer vs MLOps, data and inference engineers

This is where people get confused, so here's how I draw the lines.

The AI data engineer builds the pipelines that feed the model: ingestion, cleaning, lakehouse tables, RAG indexes. They own the data. (Our AI data engineer guide goes deep on that.)

The machine learning engineer decides what the model should learn from that data, builds and trains it, and proves it works. They own the model and its results.

The MLOps engineer builds the factory around the model: training pipelines, registries, CI/CD, monitoring and retraining automation. They own the process. (See what an MLOps engineer does.)

The inference engineer makes trained models run fast and cheaply in production, using batching, quantisation and GPU optimisation. They own serving performance. (Here's our inference engineer guide.)

In a start-up, one person might do all four. Remember "data, model, factory, speed" and you'll read job ads far more accurately.

The skills employers ask for

Skillenai indexed 2,383 postings with the title "Machine Learning Engineer" in the 90 days to 11 November 2026. The most-named skills are telling (Skillenai):

Skill	Share of postings
Python	72.0%
Machine learning	57.2%
PyTorch	49.0%
TensorFlow	31.1%
Deep learning	23.2%
AWS	17.6%
Kubernetes	16.1%
scikit-learn	15.7%
Java	15.3%
Spark	14.9%
SQL	14.3%

Source: Skillenai, 90 days ending 11 November 2026.

My read: Python plus a deep learning framework is the entry ticket, with PyTorch in almost half of all ads. What I find interesting is how much ordinary software engineering is in that list: Java, Kubernetes and Spark, with C++, CI/CD and Docker just below the cut. Employers don't want researchers who can't ship. They want engineers who understand models. In AI data engineer ads SQL appears in two-thirds of postings; here it's one in seven. The centre of gravity is the model, not the warehouse.

Beyond the keywords, what separates great ML engineers is statistical intuition, a healthy suspicion of results that look too good, and the ability to explain trade-offs to a product owner.

Two machine learning engineers reviewing model performance charts on a large screen in a bright office

Most of an ML engineer's week is spent reading results like these and deciding what to try next.

Certifications and learning paths

You don't need a certificate to become an ML engineer. But for career changers, a cloud ML certification proves you can build and deploy on the platform your target employer uses. I'd look at these first:

AWS Certified Machine Learning Engineer – Associate. AWS has retired the older Machine Learning – Specialty exam; the last day to sit it was 31 March 2026, so the Associate is now the one to aim for.
Google Cloud Professional Machine Learning Engineer
Databricks Certified Machine Learning Associate, a 90-minute, 48-question proctored exam, useful if your target employers run Databricks.

For the learning itself, combine a solid ML fundamentals course (linear models, trees, evaluation, overfitting) with a hands-on PyTorch course and lots of reading other people's code. AI assistants such as Grok Bot, Claude, ChatGPT or Gemini make useful study partners: ask them to explain a loss function or find the bug in your training loop, then make sure you understand every line.

What it pays

These sources measure different things, so I'll show the spread rather than pretend there's one magic number.

1. Our own salary benchmark. The Machine Learning Engineers salary page on AutomationRoles.ai, built from live roles, shows a US median of $220,035, with the 25th to 75th percentile running from $182,875 to $248,738, based on 174 live roles at the time of writing.

2. Posted salaries on Indeed. Indeed puts the average base salary at $185,920 a year in the US, from 5.5k salaries in job postings over the past 36 months (updated 2 November 2026).

3. Self-reported pay on Glassdoor. Glassdoor shows median total pay of $168,090, with the 25th to 75th percentile at $133,288 to $214,204 and the 90th percentile at $264,870, based on 8,748 salaries.

4. Total compensation at tech companies. Levels.fyi puts the median yearly total compensation of an ML / AI software engineer in the US at $251,000, including stock and bonus. That leans towards big tech, where equity is a big part of the package.

5. AI companies' posted ranges. The AI Hiring Index tracks posted base-salary ranges at 272 AI companies. For engineering roles, the median range is $185,000 to $265,000, while new-grad roles have a median range of $135,000 to $155,000.

6. Government statistics. The US Bureau of Labor Statistics has no dedicated ML engineer category. Its closest match, computer and information research scientists, had a median wage of $140,300 in May 2025, with the top 10% above $230,630, and employment is projected to grow 22% from 2025 to 2035, much faster than average.

7. Outside the US. In the UK, IT Jobs Watch shows a median advertised salary of £70,000 for machine learning engineer roles in the six months to 11 November 2026, up 3.7% on the same period last year.

Real ranges from roles live on AutomationRoles.ai right now: a Machine Learning Engineer, Risk & AI at OKX in San Jose lists $214,666 to $321,999; a Staff Machine Learning Engineer at Coupang in Mountain View lists $164,000 to $282,000; and a Machine Learning Engineer at Ford in Dearborn lists $74,300 to $144,900.

Putting it together, a realistic US base range for experienced ML engineers outside big tech and AI labs is roughly $140,000 to $215,000, with AI-native and big-tech employers well above that once equity counts.

Who's hiring right now

Skillenai's top employers posting the ML engineer title over the last 90 days show how broad this market is: Capital One (69 postings), ServiceNow (68), Waymo (61), Wayve (50), Reddit (46), Adobe (45), Pinterest (32) and Cisco (29). Banking, enterprise software, autonomous driving and social media are all competing for the same talent, led by San Francisco, London, New York, Mountain View and Santa Clara.

On AutomationRoles.ai, we have ML roles live at Apple, OKX, Coupang, Ford and AMD (including a Summer 2027 PhD ML engineer internship), among many others. Browse the latest jobs here.

How to break in: my practical plan

Skillenai's career-flow data shows where ML engineers come from. The most common moves into the role (apart from ML engineers switching employers) are from data scientist (418 moves), software engineer (371), research assistant (153) and data analyst (101). That tells me there are two main on-ramps, and you only need one of them.

If you're a software engineer: you already have the half most data scientists lack. Learn the ML fundamentals properly, then build one end-to-end project: take a real public dataset, train a baseline, beat it, and deploy it behind an API with simple monitoring. Write up what failed and why.

If you're a data scientist or analyst: you have the statistics. Close the engineering gap: tested Python, Git, Docker, one cloud, and getting a model out of a notebook into a service someone else can call.

For everyone: pick a domain, such as fraud, recommendations, forecasting or computer vision. Employers hire people who understand their problem, not just their tools. And put numbers on everything: "improved precision from 0.81 to 0.88 at the same recall" is worth ten bullet points of buzzwords.

My take

People ask me whether foundation models make ML engineers less important. I think the opposite is happening. Every company now wants AI that works on its data, for its customers, measured against its goals, and that's exactly the ML engineer's job. Better tools just mean a good ML engineer can ship more in a month than a whole team could a few years ago.

The bottom line: ML engineers own the model and its results. Python and PyTorch are the base, software engineering discipline is the multiplier, and honest evaluation is the superpower. If you love the moment a model finally clicks, there's never been a better time to aim for this job.