The main job families

What you will be able to do

Tell the main AI roles apart by their work, not by their titles. When you know what each role produces and how it is judged, you can choose a target, and you can read a job advertisement correctly.

Before you startWhat to know and to have readyAbout 30 minutes

What you need to know

• The five parts of an AI product, from AI in everyday products: Five parts of an AI product.
• The words AI, ML, model and inference, from AI in everyday products: Words for this course.

What you need

• Your notes.md file in the ai-career folder. You made it in AI in everyday products: Set up your course notes. If it is missing, make it again as in that section. This lesson adds the heading ## Job ads.

Judge a role by what it produces

Job titles are not consistent. One company’s “data scientist” builds dashboards. Another company’s “data scientist” trains models and puts them into production. So this lesson describes each role with three questions:

1. What does this person do on a normal day?
2. What do they produce? A role is easiest to recognise by its output.
3. How is their work judged? This tells you what the role really values.

Two more words first. Data science is the use of data, statistics and ML to answer questions and support decisions. AI engineering is building software products on top of AI models, often models that another company trained.

Six job families

Data engineer

Data engineers make data available, clean and reliable for everyone else.

• On a normal day: connect a new data source, write a pipeline that copies and cleans data every night, fix a pipeline that failed at 3 a.m.
• Produces: data pipelines and tables that other people use.
• Judged by: Is the data correct, complete and on time? Do pipelines fail, and how fast are they fixed?
• At Larkfield: collects the 40 policy documents from three places, keeps only the current version of each, and adds new versions automatically.

Data scientist

Data scientists answer questions with data, and measure whether something works.

• On a normal day: explore data, run an experiment, measure a model’s results, explain the results to people who are not technical.
• Produces: analyses, experiments and a recommendation, for example “This model is good enough for a pilot, except for refund questions.”
• Judged by: Are the conclusions correct and useful? Did a decision improve because of them?
• At Larkfield: builds a test set of past questions with correct answers, and measures how often the assistant is right.

ML engineer

ML engineers train models and make them run reliably for real users.

• On a normal day: prepare training data, train and compare models, package a model, set up monitoring.
• Produces: a trained model that runs in production, with tests and monitoring. Production is the real system that real users depend on, as opposed to an experiment on one laptop.
• Judged by: Is the model accurate, fast and stable in production? Can it be retrained safely?
• At Larkfield: not needed for the first version, because the assistant uses a language model that another company trained. If Larkfield later trains its own ticket classifier, an ML engineer builds it.

AI engineer

AI engineers build application features on top of AI models.

• On a normal day: write the instructions that the model receives, connect the model to data and tools, handle errors and wrong answers, measure cost and speed.
• Produces: an AI feature inside a product, for example the answer panel in the help desk tool.
• Judged by: Does the feature work for users, with correct answers, an acceptable cost, and a safe result when the model is wrong?
• At Larkfield: builds the assistant: the search in the documents, the instructions to the model, the links to the sources.

Solution architect

Solution architects design how all the parts fit together, before the team builds them.

• On a normal day: draw the design, compare options and their cost, check security and legal requirements, talk with every team that is affected.
• Produces: a design document and decisions, for example “The assistant runs in our cloud account and uses a model provider that we have a contract with.”
• Judged by: Does the system meet its requirements for cost, security and growth? Were expensive problems avoided early?
• At Larkfield: decides where the assistant runs, how it connects to the help desk tool, and what it may cost each month.

Researcher

Researchers invent new methods and prove that they work.

• On a normal day: read research papers, design experiments, run many experiments, write papers.
• Produces: a new method or model, with evidence that it is better.
• Judged by: Is the idea new and correct? Do other researchers use it?
• At Larkfield: none. Larkfield uses methods and models that researchers made elsewhere. Most companies are like Larkfield. Research jobs exist mainly at large technology companies, AI laboratories and universities, and many of them ask for a PhD.

How the outputs connect

Each role's output is another role's input. The architect's design shapes the whole system.

Text description of the diagram

• The data engineer’s clean data goes to the data scientist and the ML engineer.
• The data scientist’s recommendation helps the ML engineer decide what to build.
• The ML engineer’s trained model goes to the AI engineer.
• The researcher’s methods and models go to ML engineers, and through model providers, to AI engineers.
• The AI engineer’s feature goes to users.
• The solution architect’s design is the plan that the others follow.

Role	Main output	Coding	Maths and statistics
Data engineer	Data pipelines and tables	A lot: SQL and Python every day	A little
Data scientist	Analyses and recommendations	Medium: Python and SQL for analysis	A lot
ML engineer	Models in production	A lot: Python and production software	Medium to a lot
AI engineer	AI features in a product	A lot: application software	A little to medium
Solution architect	Designs and decisions	Medium: reads code, writes some	A little
Researcher	New methods, with evidence	A lot: experiment code	A lot, advanced

In small companies, one person often does two or three of these roles. In large companies, each role can be a whole team.

Guided practice: read a job advertisement

Here is a worked example. This fictional advertisement has the title “Data Scientist”:

You will build and maintain our nightly data pipelines in SQL and Python. You will make sure that the sales tables are correct and on time for the analytics team. Experience with cloud data warehouses is a plus.

Ignore the title, and look at the work: pipelines, correct tables, on time for other people. That is a data engineer job. A person who applies because they want to run experiments will be disappointed.

Now do two yourself. Write your answers in notes.md under ## Job ads.

1. Read advertisement A and advertisement B below.
2. For each one, write the job family that matches the work.
3. Write the words in the advertisement that told you.

A — “AI Specialist”: You will add an AI writing helper to our customer app. You will connect the app to a model provider, write and test the instructions for the model, and make sure the feature handles wrong or slow answers well.

B — “Machine Learning Analyst”: You will study why customers cancel their subscriptions. You will analyse our data, test which factors matter, and present your findings to the management team every month.

Show the answers

• A is an AI engineer job. The words: “add … to our customer app”, “connect the app to a model provider”, “handles wrong or slow answers”. The output is a feature in a product.
• B is a data scientist job. The words: “analyse”, “test which factors matter”, “present your findings”. The output is an analysis and a recommendation, not a system.

Common mistake

Choosing a role by its title

Symptom: you apply for jobs called “AI Engineer”, and the interviews ask about things that you did not prepare for. Or you get a job and the daily work is not what you expected.

Cause: companies use titles differently. The title tells you little. The list of tasks tells you a lot.

Fix:

1. For each advertisement, underline the verbs: build, analyse, maintain, present, design.
2. Ask: “What does this person produce?”
3. Match the output to a family in this lesson. Apply if the output is the one that you want to produce.

Check your understanding

Larkfield's assistant works well in tests, but each answer costs too much and takes 15 seconds. Which role is responsible for fixing this, and which role should have seen it earlier?

Show the answer

The AI engineer fixes it, because the cost and speed of the feature are part of their work. For example, they send less text to the model, or use a smaller model. The solution architect should have seen it earlier, because the design includes a cost limit for each month. A design that does not estimate cost is incomplete.

Check your understanding

Why does Larkfield have no researcher, and what does that tell you about most AI jobs?

Show the answer

Larkfield uses models and methods that others made. It does not need to invent a new method to solve its problem. Most companies are like this, so most AI jobs are about using AI well: data, models in production, features, design and evaluation. Research is one real path, but it is a small part of the job market.

Recap

• Judge a role by what it produces and how it is judged, not by its title.
• Data engineers produce reliable data. Data scientists produce analyses and recommendations. ML engineers produce models in production. AI engineers produce AI features. Solution architects produce designs. Researchers produce new methods.
• Each role’s output is another role’s input.
• Most companies use AI. Few invent it, so research is a small part of AI work.

Next, you meet the roles around these six: analytics, product, evaluation and governance, and how much code each one needs.

Next: Technical and adjacent routes