What is an AI engineer?

Companies like OpenAI and Anthropic serve powerful models through an API, so most teams no longer train their own from scratch. The AI engineer takes one of these models and turns it into a working product feature.

These are called foundation models, meaning large models that are trained once and then reused for many different tasks. As they moved from research labs into everyday products, someone had to connect them to real users and real data. That job is usually called AI engineer.

The title is recent. In June 2023, Shawn Wang (known online as swyx) published The Rise of the AI Engineer, an essay arguing that a new engineering specialty was forming around applying these models. He pointed out that a wide range of AI tasks that took "5 years and a research team" in 2013 "now just require API docs and a spare afternoon."

The core of the job is connecting a general-purpose model to a specific business problem. In practice that looks like:

• Designing prompts and the logic that decides what to send the model
• Building retrieval so the model can answer from a company's own documents, an approach called RAG (retrieval-augmented generation)
• Wiring model APIs into an application backend, then handling failures and rate limits
• Adding tools and function calling so the model can take actions, such as looking up an order status
• Writing test cases to check output quality, since the same prompt can return different answers
• Tracking cost and latency, because every model call takes time and money
• Shipping the feature to production and monitoring it after launch

Most of this is normal software engineering. The AI-specific part is that the model's output can change from one call to the next, so the engineer spends real time measuring quality and guarding against bad answers.

This is the most common point of confusion. The two roles sit at different layers of the same stack.

A machine learning engineer builds and trains models. They collect and prepare training data, then run the training jobs that produce a working model. This work leans on statistics and the math behind machine learning.

An AI engineer starts from a model that already exists. Their job is to make it useful inside a product by shaping what goes into the model and handling what comes out.

Chip Huyen draws the same line in the resources for her 2025 book AI Engineering. She describes AI engineering as building applications on top of foundation models, which involves more prompt engineering and context construction. Building on traditional ML models involves more feature engineering and model training.

In real job listings the titles overlap, and some roles labeled "AI engineer" still expect model training. Read the responsibilities in the posting to see which kind of work the job involves.

A data scientist analyzes data to answer business questions, often through statistics and experiments. The result is usually an analysis or a report that informs a decision.

A general software engineer builds and maintains software of many kinds. An AI engineer is a software engineer who specializes in making foundation models work reliably inside that software.

Roles vary. The usual pattern is a working software engineer who has added AI-specific skills on top, including:

• Solid general programming, usually Python and often TypeScript for web apps
• Comfort building application backends and calling external APIs
• Prompt design, plus a feel for how models behave and where they fail
• Retrieval and vector search, so a model can answer from your own data
• Evaluation, meaning test cases that measure whether the output is good enough
• An eye on cost and latency, since every production call adds up
• Enough understanding of how models are trained to reason about their limits

For what employers ask for in practice, see the skills measured across real job postings.

There is no single required degree. Many people arrive from software engineering and add AI skills, and others come from data science and add production engineering. A practical path:

1. 01Get comfortable shipping normal software first, since AI features still live inside regular apps.
2. 02Learn to call an LLM API and build something small, like a question-answering bot over your own notes.
3. 03Add retrieval so the bot can answer from your real documents.
4. 04Learn to write test cases for the output, so you can tell whether a change made things better.
5. 05Put a feature into production and watch its cost and behavior over time.

The MLGuerrilla modules teach several of these skills, including evaluation and context engineering. The applied AI engineer guide has project ideas that show you can do the work.

Is an AI engineer the same as a machine learning engineer?

No. A machine learning engineer builds and trains models. An AI engineer builds products on top of models that already exist. The titles do overlap in some listings, so check the responsibilities.

Do you need a PhD to be an AI engineer?

Usually not. The common asks are strong software engineering and hands-on experience building with models. In his 2023 essay, swyx pointed out that none of the effective AI engineers he named had a PhD. A research degree helps more for jobs that involve training models.

What programming language do AI engineers use?

Python is the usual choice, because most AI libraries and model SDKs support it well. TypeScript is common too, especially for web and product work.

Do AI engineers train their own models?

Most of the time, no. They adapt existing models through prompting and retrieval. Some also fine-tune a model on a smaller dataset. Training a foundation model from scratch is rare and expensive.