AI Engineers, Applied Scientists, Forward Deployed Engineers: Let’s Unpack.

To say that we don’t know how AI will impact our lives is to state the obvious. Between sensationalistic titles, actually impressive benchmarks, and state-of-the-art Large Language Models (LLMs) that come out every day, it is very hard to predict how the whole world is going to react, adjust, and evolve in the near future.

Of course, this also means that the job market will be impacted. In fact, it already is.

While it is normal to get scared and focus on what roles might not exist tomorrow, it is also important to see how new jobs are emerging and how existing ones are developing. This is why I wrote this.

I don’t want to give you a list of data science jobs or enumerate those differences. ChatGPT can do that very well. Instead, I’d like to show you what changed in the data science world, why new roles are coming up, and how to prepare for them.

Let’s start with the first question. How did we end up here?

0. Back to The Future

Back in the day, say before “Attention is All You Need“, the data science field was basically split in two:

1. The Data Scientist, usually closer to the product side. The job was heavily rooted in statistics, experimentation, data analysis, and extracting useful insights from data. Python and SQL were the core tools.

2. The Machine Learning Engineer, who lived much closer to the model itself. The focus was on training pipelines, optimization, neural networks, performance, deployment, and building end-to-end machine learning systems. Heavy on Python and Machine Learning libraries such as TensorFlow, PyTorch etc.

Other roles, such as Data Analyst and Data Engineer, had clearer boundaries and were easier to distinguish.

Then there was the Applied Scientist, a title popularized by companies such as Amazon and Microsoft. Applied Scientists and Data Scientists were very similar, but the Applied Scientist often sat closer to the engineering side of things: they were expected not only to experiment with new models and algorithms, but also to make sure the code could actually operate inside a production system.

The reason why the lines were quite well defined is simple: prototyping required a lot of work. To have a solid statistical analysis of a system and produce meaningful insights (Data Scientist), you needed weeks of coding, testing, debugging, and all without a coding assistant to help you.

Training a Neural Network (Machine Learning Engineer) would require setting your hyperparameter space, selecting your features, engineering them, preventing data leakage, selecting your deep learning architecture, and much more, which also takes weeks of work. To find an expert in both would have been challenging, especially because Neural Networks were not so popular and, frankly, unnecessary: the roles were somewhat connected, but also distinct.

But you can’t relax. You need to be attentive. Someone would say, “Attention is All You Need”.

1. Attention, Please.

Attention Is All You Need is a paper developed by Google in 2017. The paper is a game-changer because it introduced the transformer architecture, which is the base algorithm for LLMs like GPT.

It’s been a wild ride since then: GPT, GPT-2, GPT-3, GPT-4, GPT-5, all the way to GPT-6 Astra (as of October 3, 2026).

People in the field had already started paying serious attention to transformers with GPT-3. However, GPT -4 was the real game changer: companies realized they could start building real products around it.

Somewhere between GPT-4 and GPT-5, a new role started appearing everywhere: the Artificial Intelligence Engineer, or AI Engineer.

The AI Engineer is a whole new role.

This person understands Large Language Models and knows how to use them. They know how to design and tune prompts, choose the right model for a given task, build guardrails, anticipate hallucinations, evaluate outputs, and make LLM-based systems reliable enough to be used in production.

An AI Engineer might come from a data science background, but they might not. They might have trained a machine learning model before, but they might not. While these skills are certainly useful, as they are evidence of a solid technical foundation, they don’t necessarily define the role: when companies hire AI Engineers, that is not the evidence they are looking for.

They want to see agents built, LLM wrappers shipped, and LLM workflows running in production.

Ok, so we have a new role, cool. Why would Data Scientists (or Machine Learning Engineers) care?

2. It’s Alive!

The most capable LLMs are not so powerful (just) because they have a trained Neural Network that is optimized correctly. That is just the starting point.

What you see packaged as ChatGPT or Codex or Claude is really an orchestrated system of many LLMs interacting with each other. When this level of complexity was achieved, another evolution of AI was unlocked: agentic AI brought automation for productivity.

Think about 2023/2024. Maybe you used ChatGPT for some of your tasks: writing an email, fixing some bugs, etc. However, that was not really “automation” per se. You were just using a productivity tool to speed up some parts of your pipelines.

Tools like Codex, Cursor, or modern ChatGPT plugins are completely different. They automatically check email for you; they open MRs; they check for bugs; they write entire codebases for your projects. They perform “actions,” which is why we call them “agentic”.

They are also very good at doing actions, especially in programming. This means that, for Data Scientists, coding is not the barrier anymore.

This changes both the Machine Learning and the Data Scientist roles, and how companies are looking at them. Modern DSes and MLEs are closer to the Product Manager part of the pipelines than they were before, because, with coding being more automated, they have more time to find problems, design systems, and test them.

A good DS also doesn’t create bottlenecks in the pipeline by just developing sloppy prototypes that engineers have to fix. This means that they are also expected to have a bit of software engineering in them as well, to make Software Engineers’ lives easier.

In short, DSes are required to be hybrid figures. There is also one hybrid figure that is hybrid on steroids. Let’s talk about it.

3. You Can Get With This, or You Can Get With That

The Forward Deployed Engineer term was first introduced by Palantir. Palantir was working with clients that had complex and somewhat open problems. In order to design solutions, engineers needed to directly interact with customers to define the issues in detail and think of an optimal solution.

Palantir’s story tells us the origin of this word. The way we use it in the AI world is quite different.

As AI becomes more popular, companies look at it as a possible solution for their problems. Similar to what happened with Palantir, they might not be familiar with the technology. For this reason, they need a person who can actually speak to customers, understand the problem in depth, and then create a ready-to-execute design to solve it.

In the AI world, this figure is the most hybrid you can think of. They need some of the technical depth of an AI Engineer and/or Machine Learning Engineer, some of the analytical thinking of a Data Scientist, and some of the product intuition of a Product Manager.

A Forward Deployed Engineer understands the technology well enough to build the solution, the customer well enough to identify the right problem, and the business well enough to understand whether that solution is actually creating value.

4. Cut to the Chase!

So this is how things changed. I understand that a person who studied computer science or statistics is not looking at this article like it is a History Channel show, but they are asking themselves a very practical question:

“How do I stay relevant?”

And this is a million-dollar question to which I won’t pretend to have an objective answer. I will just share my recipe, which is extremely subjective, and is the following:

1. Don’t rely on your coding skills. Sure, AI makes mistakes, and you need to be able to spot them, but that is not an edge, especially as the LLMs get sharper and sharper.

2. Become a design fanatic. Look at best practices; challenge the first design that you think of; optimize the metrics. Always be on the quest for the best solution.

3. Parallelize your work. In the era of AI, companies expect quality AND speed. AI automation allows you to parallelize multiple projects, or parts of the same one, with more agility. Use this superpower.

4. Understand the business side of the projects. This was already true; AI just made it more obvious. With coding not being a barrier, Data Scientists can spend more time defining a quantitative way of measuring the quality of their algorithms and the impact they produce.

5. Find your angle. AI-induced panic might make you want to jump ship. Knowing a little bit about everything is helpful, but if you sell yourself as everything at once, people will struggle to see you as an expert in anything. Don’t forget to keep your own credibility in a specific field.

And, as always, keep it easy. Change is expected, but your skills are not going anywhere.

5. Conclusions

Thank you for making it all the way here. It means a lot. ❤️

Here is what we covered in this article:

• We reflected on how the data science world used to be and the original sharp distinctions between Data Scientists, Machine Learning Engineers, and Applied Scientists.

• We saw how Transformers and LLMs changed that landscape, eventually giving rise to AI Engineers.

• We discussed how agentic AI changed Data Scientists’ day-to-day work, making coding less of a bottleneck and pushing technical roles closer to product and system design.

• We looked at the Forward Deployed Engineer as one of the clearest examples of this new hybrid way of working.

• Finally, we talked about what all of this means for Data Scientists and Machine Learning Engineers who want to stay relevant as AI keeps improving.

Before you head out!

Thank you again for your time. It means a lot. My name is Piero Paialunga, and I’m this guy here:

I’m originally from Italy, hold a Ph.D. from the University of Cincinnati, and work as a Data Scientist at The Trade Desk in New York City. I write about AI, Machine Learning, and the evolving role of data scientists both here on TDS and on LinkedIn. If you liked the article and want to know more about machine learning and follow my studies, you can:

A. Follow me on LinkedIn, where I publish all my stories
B. Follow me on GitHub, where you can see all my code
C. For questions, you can send me an email at piero.paialunga@hotmail.com