The AI Engineering Stack
AI engineering is the field that has grown from ML engineering, and it brings new challenges and solutions alongside its older roots.
Why the Field Feels Overwhelming
AI engineering is moving quickly. New techniques, discoveries, and engineering achievements appear almost every day, and a large community is contributing creativity and engineering talent at a pace that is hard to describe. The energy around the field is real and broad, but the sheer volume of activity can feel overwhelming to anyone trying to understand it.
A natural reaction is to hope that tools will solve the problem. AI tools are good at gathering information from many places and condensing it, so they can help summarize new developments. Tools only go so far, however. The more crowded a space becomes, the more useful it is to have a framework that tells you what matters and how the pieces relate to one another. The framework used in this material starts from the component everything else depends on: the foundation model.
In a crowded field, a framework that organizes the pieces is more valuable than relying on tools to aggregate every update. Understanding the layers comes first; the details fill in around them.
Two Eras of Machine Learning Work
Traditional ML refers to all machine learning done before the advent of foundation models. AI engineering is the field that has grown out of ML engineering; it is built on foundation models and brings new challenges and solutions alongside its older roots.
AI engineering is not simply a new name for the same work. The engineering stack has changed from what ML engineering looked like, and that change brings new challenges along with new solutions. The useful first step is seeing the difference clearly, because that difference explains why the rest of the framework is organized the way it is.
Read the comparison column by column. The left column collects the older practice, every form of ML that came before foundation models. The right column describes the field taking shape now. The arrow marks the change in the engineering stack, which is the central point. The comparison does not claim the two columns share no ideas; the older roots remain, and the comparison highlights where the emphasis has moved.
| Aspect | Traditional ML (before foundation models) | AI engineering (built on foundation models) |
|---|---|---|
| Scope | All machine learning done before foundation models arrived | The field that has grown from ML engineering, with foundation models as its base |
| Relationship to older work | The original practice | Grown out of ML engineering, keeping its older roots |
| Challenges | The challenges of earlier ML work | Challenges carried over from ML engineering plus new ones |
| Central building block | Not defined by foundation models | The foundation model, the fundamental building block |
A comparison drawn only from the framing of the source material. Specific workflow details are not stated in the source and are not listed here.
The Stack as a Layered Dependency
A foundation model is the fundamental building block of AI engineering, and it is what makes many amazing applications possible. The most useful way to think about it is as a base layer. Applications are built on top of it, and the framework of this course starts from that base layer before moving upward.
The stack reads from the bottom up as a dependency. AI engineering begins with foundation models, and foundation models are what make the many applications at the top possible. Because the arrangement is layered, a change at the foundation level affects everything resting on it. This is the reason the course gives the foundation layer so much attention.
Tracing a Foundation Model into an Application
Now trace the path a foundation model takes to become part of an application a user interacts with. The source gives the layered view but does not spell out implementation steps, so the trace below follows the base-to-top ordering the source describes. Each step is a layer, and the arrows show where the dependency flows.
Tracing one foundation model through the stack
A pretrained foundation model is available. Trace how it becomes part of an application that a user interacts with, using only the layered view.
- 1Start at the base
The foundation model is the fundamental building block of AI engineering. It is the bottom layer, the thing everything else depends on.
- 2Identify what rests on it
Applications are built on top of the foundation model. The base layer is what makes those applications possible, so the application layer depends on the foundation layer.
- 3Follow the dependency upward
The stack reads from the bottom up as a dependency. The foundation model feeds the layers above it, which means the path from base to user runs upward through the stack.
- 4Locate the user
The user meets the application at the top of the stack. The user never interacts with the base layer directly in this view, but the application's behavior ultimately rests on it.
- 5Predict the effect of a base change
Because the arrangement is layered, a change at the foundation level affects everything resting on it, including the applications the user sees.
The foundation model becomes part of the application by serving as the base layer that applications are built on. Every layer above it depends on that base, so a change at the bottom propagates upward.
Mistakes Learners Make with the Layered View
Common mistake Treating AI engineering as just a new label for ML engineering
What it looks likeAssuming the work is the same because the field grew from ML engineering.
Why it failsThe source says the engineering stack has changed from what ML engineering looked like, and the change brings new challenges along with new solutions.
Do this insteadRecognize the continuity with ML engineering roots, and also the shift in emphasis that the foundation model introduces.
Common mistake Reading the diagram as claiming the two eras share no ideas
What it looks likeConcluding that traditional ML knowledge is useless once foundation models appear.
Why it failsThe comparison highlights where emphasis has shifted; it explicitly does not claim the columns share no ideas.
Do this insteadKeep the older roots in view while noting where the emphasis has moved.
Common mistake Thinking a framework is unnecessary because tools can summarize developments
What it looks likeRelying only on an AI tool to condense new work and assuming that covers what matters.
Why it failsTools only go so far. In a crowded space, a framework showing what matters and how the pieces relate is more valuable.
Do this insteadUse tools for gathering information, but use a layered framework to decide what matters.
Common mistake Assuming a change at the top of the stack stays at the top
What it looks likeBelieving that an edit to an application leaves the foundation layer untouched, and that a foundation change affects only one application.
Why it failsThe stack is a layered dependency: a change at the foundation level affects everything resting on it.
Do this insteadTrace dependencies downward to the base layer when analyzing how a change will propagate.
Checking Your Layered Reasoning
What do you think happens?
If the foundation model at the base of the stack changes, which parts of the system are affected?
Reveal answer
The foundation model and everything resting on it
The stack is layered, and a change at the foundation level affects everything resting on it. This is why the course gives the foundation layer so much attention.
Without looking back, explain in your own words the difference between traditional ML and AI engineering built on foundation models. Then describe what sits beneath a user-facing application and why a change at the bottom of that stack reaches the top.
Hints
- Start with the definition of traditional ML as all ML done before foundation models.
- Think about which layer is the base and which layers rest on it.
- The stack reads from the bottom up as a dependency.
Key Takeaways
- AI engineering is the field that has grown from ML engineering. It is not a new name for the same work: it carries older roots forward while bringing new challenges and solutions.
- Traditional ML refers to all machine learning done before foundation models. AI engineering is built on foundation models, which are its fundamental building block.
- The stack is layered. Applications sit on top of a foundation model, and a change at the foundation level affects everything resting on it.
- In a crowded field, a framework that shows what matters and how the pieces relate is more valuable than relying on tools to aggregate every update.