Nowadays, you can build almost everything with AI. With the vast development and enhancement from the AI models, now it is harder to catch up for what to develop with AI as now there is almost broad possibilities with what AI can do. You can make games, make songs, great videos with narration and voices, and make so much more. I even made Kubernetes system with the help of AI. With more variation of AI models, with specific job to do (i.e. AI song models only configured to make songs), the results have been incredible.
But now you wonder, what would you want to build with AI? If I want to make games, thousands of people nowadays uses AI to make games. If I want to make videos, there are millions of AI videos on Youtube. The AI-generated content become much less valuable. They also included botting so they generate the content constantly without the touch of human interaction. You will attract less viewer than bot generator user.
What To Build?
Well for starter, you can build some web application. It is a good start of "vibe coding" and you will learn a lot of programming terms. You will learn about html files, php script, css, js, vite, and so much more. You will also learn about file structure like xml or json. It is a good start of development. Try to grasp the basic prompting of the AI, and you're good to go.
But for more experienced developer, you might find making basic apps with AI so boring. It only repeats of what you did everyday, but just make the app so much better. And then that's it, no new knowledge or experience. You might ended up quitting your vibe coding.
There are broad possibilities of what AI can do. But is the AI can do manual tasks like human do? For example can AI play games? Diagnoses metrics and make decisions? Uses natural language as an analyst? YES, they can!
It is called AI Engineering.
What is AI Engineering
There are tons of AI models out there, like Opus, Fable, Gemini Flash, etc. Basically, they can do what they were told to do. It is called prompting. There are two type of prompt, basic prompt and system prompt. Basic prompts are the sentences we wrote to the chatbox to tell AI what to do. System prompts are the injected sentences which was told to the AI models, of the DOs and DON'Ts, as a safeguard to the AI models. For example they can modify the source code of our programs, but they can't publish our credentials to the repository.
And also there are network system within the AI. The AI have databases of which they can access to, and which ones also they don't. We can also safeguard the AI to disable the access to the sensitive information of which we don't need to access with AI to limit the possibilities of AI interaction with the databases.
This is the fun part. How we do "engineering" of the AI, so it only do what it was told to do, and doesn't do what it did not told to do. So basically not to hallucinate. We made a system prompt. Something goes like this:
You are an autonomous AI Game Player. Your objective is to play a turn-based RPG game, make intelligent decisions, survive battles, and defeat enemies. You are NOT a game assistant. You are the PLAYER controlling the character. GAME RULES: - The game is turn-based. - You receive the current game state as JSON. - You must select exactly ONE action per turn. - You cannot perform actions outside the allowed actions. AVAILABLE ACTIONS: 1. ATTACK - Deal damage to the enemy. 2. DEFEND - Reduce incoming enemy damage. 3. HEAL - Restore your HP using a potion. DECISION PRIORITIES: 1. If the enemy can be defeated with one attack, choose ATTACK. 2. If your HP is critically low (below 35%) and you have potions, choose HEAL. 3. If the enemy is preparing a powerful attack, choose DEFEND. 4. Otherwise, choose ATTACK. RESTRICTIONS: - Never invent game state information. - Never choose an unavailable action. - Never use HEAL if potions are zero. - Do not ask the user what to do. - Do not explain your decisions outside JSON. - Base decisions only on the provided game state. OUTPUT FORMAT: Return ONLY valid JSON: { "action": "ATTACK", "reason": "Brief explanation of the decision" } Your goal is to maximize survival and win the game. The AI do as it was told to do, and do not do other things other than its responsibilities. It was simple prompting, but this was the important part of AI engineering.
Also we need to manage of what the AI have access to. For example the AI only have direct access to the game window, then it only input the command control to the current window, and do not have access to other open window. And which databases it will have access to to keep the learning data to be read.
I think those are fun part of an AI Engineering projects. Of which will have trial and error part. I guarantee you this will be a fun experience :)

