AI writes the code. I still make the call.

AI and Expertise

AI still lacks the concepts internalized by people who are experts in their own fields. I ran into this idea today while reading Scott Young’s text “Is Vibe Coding the Future of Skilled Work?” He talks about things he “vibe coded” but needed the knowledge to do that, so he created systems that would not work without that knowledge.

Floor and Ceiling

Scott says that right now, with AI and vibe coding, two things happen at once. The floor to develop a project drops, but the ceiling to develop and maintain a good project rises. Building something that runs is easier than ever, you can go to any AI, like ChatGPT, prompt a few things, and get a prototype or even a minimum viable product that shows a bit of the solution you want.

But building something that holds now has a higher bar. Because everyone can make anything simple, the demand for truly professional, high‑quality systems rises. Shitty versions of code are on the rise, and so are the good, professional versions. The hard part moved; it didn’t disappear. It’s now harder to build something that isn’t just an MVP or a prototype.

Prototype vs Production

In my daily work, I see clients get a prototype very fast, but they still have no way to finish it because they lack the technical knowledge to run the project. The idea is cheap, but the knowledge didn’t come with it. AI can write code, but when I create a project I still have to make the calls. I get code that looks correct, yet it isn’t quite right. I still need to think about how to test, validate, and check whether it really solves the problem.

Testing Challenges

For example, a number might add up incorrectly in a financial system I built. AI doesn’t get all the context at once, so I have to plan carefully and explain everything to the AI. If you ask AI to write tests, it often produces useless tests. That’s another problem: AI can implement something, but it doesn’t propose the right testing strategy.

Domain Knowledge

What AI lacks are the concepts an expert carries. AI can implement, but it doesn’t make the high‑level decisions. You still need an expert to plan the project and then delegate the writing part to AI. AI is like a typewriter on asteroids, it does a lot faster than a human, but the decision‑making stays with the human.

The human still needs domain knowledge, and you can’t prompt your way into that. If you don’t know, say, how Elementor in WordPress integrates with Claude Code, you’ll hit a wall. I tried to get Claude to reproduce an Elementor page in code with a single prompt, and it failed. I had to explain how Elementor structures the page because the model didn’t understand the core domain knowledge.

Importance of Fundamentals

That’s why studying the fundamentals is more important than ever. Understanding the theory of your field, not just the tool you’re using, builds the domain knowledge you need. If you’re not technical and you’ve already shipped a project, put someone technical on it. A project built by AI is not the same as a project that’s been tested.

Safety for Non‑Technical

Last week a client asked me to code a simple system to manage his leads. I discovered an API that could be accessed without authentication, a basic security flaw. I had to fix it myself because the client didn’t have any cyber‑security knowledge. If you’re not technical, please have someone test your project to avoid regrets from using AI.

Final Thought

The main idea from today’s video is that code has gotten cheap, but knowing what you’re doing has not. I care a lot about being good at my work, not for vanity but because it’s owed to whoever I serve. I draw this thinking from Catholic examples, especially Saint Josemaría, whose text I’ll link in the description. See you in the next video.