AI is everywhere now. Every week there is a new app, a new tool, or a new idea. But what are people actually building with AI? And how should you plan an AI app so it does not fail?

In this post, I share what I learned from the book AI Engineering by Chip Huyen. I cover the main foundation model use cases, how to plan AI applications, and the role of AI and humans in an app. I wrote it in simple words, so anyone can understand it, even if you are new to AI.

What Is a Foundation Model?

A foundation model is a large AI model trained on a huge amount of data. Because it learns from so much data, it can do many different jobs, like writing, coding, answering questions, and making images. You do not need to build a new model for each job. You can use one strong model and shape it for what you need.

This is why so many people are building AI applications today. The book groups these applications into a few main types. Seeing them in groups helped me understand the big picture.

The Main Foundation Model Use Cases

1. Coding

Coding is one of the most popular uses of AI. AI can write code, explain code, find bugs, and help you finish work faster. For students and developers, this is very useful. But you still need to understand the code yourself, because AI can make mistakes.

2. Image and Video Production

AI can create pictures, edit photos, and help make videos. This is used in design, ads, social media, and creative work. It saves a lot of time, and it lets people without design skills make good-looking content.

3. Writing

AI helps people write emails, reports, articles, and ads. It can also fix grammar and improve text you already wrote. A lot of people use it every day without even thinking about it.

4. Education

AI can be like a personal teacher. It can explain a hard topic in an easy way, make practice questions, and let you learn at your own speed. Every student learns differently, so this is a big help. It works best when you use it to learn, not just to copy answers.

5. Conversational Bots

These are chatbots. They answer questions, help customers, and talk with people. Some are made for customer support. Some are made to be a friend or a coach. This is one of the most common uses of AI.

6. Information Aggregation

Today we have too much information. Emails, meetings, articles, and messages never stop. AI can read a lot of it and give you a short summary. This saves time and helps you focus on what matters.

7. Data Organization

Many companies have messy data, like documents, images, and notes. AI can sort and organize this data, so it is easier to search and use. This is a quiet use case, but it is very useful for businesses.

8. Workflow Automation

AI can do boring and repeated tasks for you. It can also connect many steps of a job and do them one by one, like booking, sending, and updating. When AI can plan and take actions on its own, it becomes much more powerful.

Internal Apps Come First

One thing I noticed is that many companies build internal apps before public apps. Internal apps are used by their own team. Public apps are used by customers. Internal apps are safer because the risk is lower. If something goes wrong, it does not hurt customers or the company’s name. It is a smart way to start.

Planning an AI Application

Here is the most important lesson for me: building a cool demo is easy, but building a good product is hard. AI apps can look amazing in a demo, but real users bring real problems. So you must plan before you build.

The first question to ask is: why am I building this? The book gives a simple way to think about it:

• Is it a must-do? If your competitors use AI and you do not, you might fall behind. Then AI is a matter of survival.
• Is it a big chance? AI might help you earn more or save money, even if nobody forces you.
• Are you not sure? You may want to try AI so you do not miss out later, but you do not know the value yet.

Your answer matters because it decides how much time and money you should put in. If AI is a must-do, you invest more. If you are only exploring, start small.

The Role of AI and Humans in an App

This part was my favorite. When you build an AI app, you must decide how AI and humans will work together. Here are the key questions.

Is AI Critical or Complementary?

If the app cannot work without AI, then AI is critical. If the app still works without AI, and AI only makes it better, then AI is complementary.

Complementary is usually safer. If the AI makes a mistake, the app still works and the damage is small. Critical AI needs to be much more accurate, because the whole app depends on it.

Is AI Reactive or Proactive?

A reactive feature answers when you ask for it, like a chatbot. A proactive feature gives help without you asking, like a useful alert or suggestion.

Proactive features can feel great when they are right. But when they are wrong, they annoy people. Reactive features give users more control, so they are usually safer to start with.

How Much Should Humans Be Involved?

The book talks about a “crawl, walk, run” way of growing:

1. Crawl: Humans are involved in every step. AI only helps.
2. Walk: AI works directly with your own team, and humans check the results.
3. Run: When you trust the AI more, it works with outside users with less human help.

This is a careful way to build trust. You do not give AI full freedom on day one. You give it more freedom when it proves it is good.

My Main Takeaway

AI is powerful, but it should not be rushed. These are the things I will remember:

• AI has many use cases, from coding and writing to education and automation.
• Know your real reason before you build anything.
• Start small and safe, with internal or complementary features.
• Keep a human in the loop, and give AI more freedom slowly.

If you are a student, a developer, or just curious about AI, I really recommend reading AI Engineering by Chip Huyen. These are only my notes, and the book explains everything in much more detail.

What do you think is the best use of AI today? Share your thoughts, and I will share more of what I learn from this book soon.