What Does an AI Engineer Do? The Three Jobs Behind AI Apps
An AI engineer takes a ready-made AI model and builds a real product around it. One small lunch-box business shows the three jobs that turn a guessing chat assistant into one you can trust.
The lunch that wasn't paused
Bob is going away for a week. They message their lunch-box service's chat assistant: "Can you pause my tiffin next week? I'm away." The assistant replies at once: "Done! Your lunch is paused for next week."
On Monday at 12:30, the lunch box arrives anyway. Nobody is home. Nothing was ever paused.
By the end of this lesson you'll know what an AI engineer does, and the three jobs that would have saved that lunch.
Four questions, one right answer
Alex runs a small tiffin service: home-cooked lunch boxes, delivered every weekday. Customers kept asking the same things, so Alex connected a ready-made AI model, the kind behind chat assistants, to the shop's chat. It took one afternoon. Alex felt like a genius.
Then Alex tried four real customer questions. Before you read the table: how many do you think it got right?
| Customer asks | Model says | Truth |
|---|---|---|
| How do I reheat it? | Covered, two minutes | Right: general knowledge |
| What's on today? | Paneer | Wrong: today is dal and rice |
| Deliver to Sector 9? | Yes | Wrong: riders don't go that far |
| Pause next week? | Done | Wrong: nothing changed |
How do I reheat it?
- Model says
- Covered, two minutes
- Truth
- Right: general knowledge
What's on today?
- Model says
- Paneer
- Truth
- Wrong: today is dal and rice
Deliver to Sector 9?
- Model says
- Yes
- Truth
- Wrong: riders don't go that far
Pause next week?
- Model says
- Done
- Truth
- Wrong: nothing changed
One right out of four. But the model isn't broken. It knows food and polite English. It doesn't know Alex's kitchen, and it can't do anything: it only writes words.
The three jobs around the model
Look at the parts. Bob's message doesn't go straight to the model; it reaches Alex's chat app. That app is the part someone has to build, and it has three jobs.
- Bob (customer) to Alex's chat app (the part you build)
- Menu and area (the shop's facts) to Alex's chat app (the part you build)
- Alex's chat app (the part you build) to Ready-made AI model (writes the reply)
- Alex's chat app (the part you build) to Order book (real actions)
- The facts. Before asking the model anything, the app fetches today's menu and the delivery area, and sends them along with Bob's question. Now the model answers from Alex's real menu, not a guess.
- Real actions. Words alone can't pause a lunch. The model can only write a request, like "pause Bob's lunches next week"; the app reads that request and makes the change in the order book. Only when the order book says paused does the assistant tell Bob "done".
- Tests. A change that fixes one answer can break another, so Alex runs the same four questions again after every change, and each new customer question becomes a new test.
| Version | Right out of four | What changed |
|---|---|---|
| Model alone | 1 | Only general knowledge works |
| Plus the facts | 3 | Menu and Sector 9 fixed |
| Plus the order book | 4 | The pause really happens |
Model alone
- Right out of four
- 1
- What changed
- Only general knowledge works
Plus the facts
- Right out of four
- 3
- What changed
- Menu and Sector 9 fixed
Plus the order book
- Right out of four
- 4
- What changed
- The pause really happens
So what is an AI engineer?
Facts, actions, tests. The person who does those three jobs is called an AI engineer: someone who takes a ready-made AI model and builds a real product around it.
The models behind today's chat assistants are large language models: programs trained on huge amounts of text to write the next words of a reply. They're one kind of AI, and the kind most AI engineers build with today. You'll see how they work later in the roadmap; for now, a model that writes words is enough.
Two job titles sit close to this one. An ML engineer is closer to the model itself: training, tuning and serving it. A data scientist studies data to answer questions and measure what works. The next lesson puts the three side by side.
The common mix-up
Many people picture an AI engineer inventing the AI. That's the myth. Training a big model from scratch means feeding it huge amounts of text, which takes powerful computers, and a small number of companies do most of it.
Most AI engineers start from a model someone else trained. They build what goes around it, with everyday software skills and a lot of testing. If you can write a program that reads from a database and calls a web service, you already have the base.
Back to Bob
Bob asks the same question again. This time the app checks the order book first, makes the change, and the assistant replies: "Paused, Monday to Friday. See you the week after!" It really is paused. On Monday, nobody knocks.
So remember. An AI engineer takes a ready-made AI model and builds the facts, actions and tests around it.
Summary
Key takeaways
- An AI engineer takes a ready-made AI model and builds the facts, actions and tests around it.
- A model on its own knows general things, not your business: Alex's got one of four questions right.
- The model only writes words; the app makes real changes, like pausing an order.
- Running the same questions after every change shows whether a fix worked.
- Most AI engineers don't train models; they build products with everyday software skills.
Interview corner
What does an AI engineer do?
They take an existing AI model and build a product around it: fetch the facts it needs, let it take real actions through the app, and test it on real questions after every change. Most don't train the model themselves.
Practice
Check your understanding
0 of 4 answered
Finished this lesson?
Mark it done to track your progress. Saved in this browser only.
FAQ
Frequently asked questions
What does an AI engineer do day to day?
They build products on top of existing AI models: connecting the model to the facts it needs, to the actions it may take, and testing its answers on real questions after every change.
Do AI engineers build AI models?
Most don't. Training a big model takes huge data and computing power, and a few companies do most of it. AI engineers usually start from a model someone else trained.
Do I need a maths degree to become an AI engineer?
No. The core of the job is everyday software work: reading data, calling services, and testing. Maths helps more if you later move closer to training models.
Sources
