Generative vs Predictive AI
Level 3 · Kinds of AI
The three jobs AI does today. Predictive AI sorts and scores, generative AI makes new things, and agentic AI takes steps to finish a task.
Course contentsLesson 6 of 10
Level 1 · What AI is
Level 2 · AI built from rules
Level 3 · Kinds of AI
Level 4 · AI history
Level 5 · Big questions
After this lesson you can
- Tell generative AI from predictive AI
- Say what makes AI agentic
- Pick the right kind for a job
12 mincovers 3 glossary entries
Worth reading first: Narrow AI vs General AI (AGI)
Generative, predictive or agentic: which job each one does
Predictive AI sorts, scores or forecasts. Generative AI makes something new, such as text, a picture or code. Agentic AI takes steps and uses tools to finish a task. Most AI in business before 2022 was predictive. Since ChatGPT came out in November 2022, generative AI has been the kind in the news.
Predictive AI looks at data and hands back a label or a number. Is this email spam? How many coats will the shop sell next week? Will this loan be paid back? The answer is short, and you can check it against what really happens. That makes it cheap to run and easy to trust.
Generative AI learns the patterns in huge piles of real examples. Then it makes new things that look like them. You ask in a few plain words, and it writes a poem or draws a red fox reading a book. Chat apps and picture makers are both generative AI.
Its output is long and new each time, so it is harder to check. It can write a fact that sounds right and is false. It also costs more to run, because a page of text takes far more work than a yes or a no. Check what it makes before you use it.
Agentic AI is built on top of a generative model. You give it a goal, not a question. It plans the steps, uses tools such as a search box or an online form, checks its own work and keeps going until the task is done. A chat app that only writes text back is not agentic.
Agentic AI saves the most time and has the most risk. It acts in the world, so a mistake is a booking made or an email sent, not just a wrong line. Good agentic apps ask the person to say yes before any step that costs money or cannot be undone.
The common mistake is to reach for a chat app for every job. Many jobs need only a yes, a no or a number. A small predictive model does those faster, cheaper and with fewer surprises. Use generative AI when you need something new. Use a predictive model when you need a choice made.
The three kinds side by side
| Kind | What it gives back | Example |
|---|---|---|
| Predictive AI | A label or a number | Spam or not spam |
| Generative AI | New text, pictures, sound or code | A poem or a drawing |
| Agentic AI | Steps taken towards a goal | A trip searched and booked |
Which one fits
- 1
Pick predictive AI
When you need a yes, a no or a number that you can check later.
- 2
Pick generative AI
When you want a first draft or a picture, and a person reads it before it is used.
- 3
Pick agentic AI
When a task takes many steps across tools, and you can confirm each step that costs money.
Worked example
Lucy's shop uses all three
Lucy's shop uses a predictive model to guess next week's orders, so she buys the right stock. A chat app writes the first draft of each new product page.
She tried an agentic app that reorders stock by itself. Now it drafts each order, and she clicks to confirm anything over 500 euros.
The words for this lesson3
Check yourself
Which kind of AI is a spam filter?
Finished this lesson?
Marking it complete also marks its 3 glossary entries learned, which updates your roadmap and your saved page. 0 are already marked.