Generative AI: the big picture
What makes AI "generative," the five modalities, why output is probabilistic, the four-layer product stack, and build vs. buy vs. fine-tune.
Buy the model and build the system around it, for jobs that are slow by hand, cheap to check and safe to sometimes get wrong.
The first module of the Generative AI family.
Generative AI is software that creates new content instead of only analyzing existing content. A traditional model scores, ranks, or sorts. A generative model writes an email, draws an image, or writes code. This shift is not a small feature upgrade. It changes what software can do, and it changes how software fails.
This module is the map. It teaches what makes a model "generative," the five modalities it can work in, why generative output is probabilistic and what that costs a product team, where generative AI fits in your stack, when to build versus buy versus fine-tune, and where the technology creates real value and where it quietly destroys it. Every later module in this family — LLMs, RAG, agents, evaluation, security, cost — is a deeper look at one piece of the picture this module draws first.
The knowledge graph
Generative AI is not one technology. It is a shift in what software does, with consequences that spread through the whole stack. Every lesson in this module hangs off this picture:
Five lessons · shift → modalities → cost → stack → value
Predictive AI judges. Generative AI makes. The difference reshapes everything downstream.
Predictive vs generative — different jobs, not different sizes
Predictive AI
Scores · ranks · classifies
Generative AI
Creates new content
Five modalities · each with its own cost, latency, risk shape
Probabilistic software · same input, different answers
Variance
Output distribution, not one answer
New failure shapes
Hallucination · drift
Four layers · every real product wires all four
Read it in three passes. The shift: generative AI is a different job from predictive AI, not a bigger version of it. The cost: because output is probabilistic, every system built on it inherits a new class of failure — and that risk is what the rest of the family exists to manage. The stack and the call: the model sits inside a larger system — grounding, action, and operations — and every product leader has to decide how much of that system to build, buy, or fine-tune, knowing where the payoff and the danger both live.
The lessons
- What makes AI "generative"? — predictive versus generative AI, and the test for which one your product actually needs.
- The five modalities — text, image, audio, video, and code, and what changes when a model creates each one.
- Probabilistic software — what it costs a product team when the same input can produce a different output.
- The generative AI product stack — a map from the model to the running product, and where each later module in this family fits.
- Build, buy, or fine-tune — the three ways to get generative AI into your product, and how to choose at altitude.
- Where generative AI creates and destroys value — the honest pattern behind the wins and the write-offs.
Each lesson pairs the mechanics with a 🎯 For the product leader briefing — why it matters, the decision it changes, the question to ask your team, and the risk if ignored — plus a diagram. Where a lesson touches deeper mechanics, it links to the module that covers them in full: RAG & vector databases for grounding a model in real data, and Agentic AI for the loop that turns a model into something that acts.
📌 Close out the module: Recap & real-world examples.