Lesson 2 of 6
The four flavors
7 min read
'AI product' covers a fraud detector, a chatbot, a trip-booking agent, and Gmail's Smart Reply. Those feel nothing alike. How many kinds are there, really?
Four shapes, one map
Under the umbrella, most AI products are one of four flavors — and knowing which you're building sets your data, your risks, and your UX. Predictive scores or forecasts. Generative makes new content. Agentic plans and acts across steps. Embedded hides AI inside a product you already use. Tap through the map and see what each one actually learns.
Four flavors: predictive (score it), generative (make it), agentic (do it), embedded (hide it). Most real products mix them — a generative core with an agentic layer, embedded in an existing app.
Why the flavor decides everything
The flavor isn't just a label — it's your whole plan. Predictive lives or dies on labeled historical data. Generative needs evaluation and guardrails because the output is open-ended. Agentic adds tools, permissions, and the risk of compounding mistakes. Embedded is a distribution play — the AI is invisible, so the only bar is 'does the feature feel better?' Pick the flavor consciously; don't back into it.
Each flavor front-loads a different hard problem: predictive → data, generative → evaluation, agentic → safe actions, embedded → seamless UX. Name yours early.
The shape of it
- —Predictive — learns from labeled history to score or forecast.
- —Generative — learns the shape of content to make new content.
- —Agentic — plans multi-step actions and uses tools to reach a goal.
- —Embedded — a model tucked invisibly inside an existing product.
You're adding a feature that drafts a reply and, if you approve, sends it and files the ticket. Which flavor is that?
Continue in the app
Take the whole AI Product Strategy course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.