In the early days, using LLMs was simple.

You only had to decide who to ask: ChatGPT, Claude, Gemini, or maybe all of them.

Then agents and harnesses came along, and we had to think about which harness should use which model.

⚙️ Then things like effort and thinking were added. Now we also have to choose between low, medium, and high for each task, especially because these choices can directly affect the cost.

We are entering a stage where the main question is no longer just:

“Which model is stronger?”

The better question is:

“Which combination works best for this task?”

🔍 This means we need to understand these layers better: how they work, what they are good for, and when they are useful.

For example, should we set effort to high to get a better result, or should we use a specific skill instead?

For the task we want to do, which harness, which model, which plugin, and which level of access are the right choices?

🧩 When these parts are combined well, sometimes we can get a very good result from a model that is not even top-tier.

But when they are combined badly, we mostly get more latency, more cost, and a stronger feeling that the answer is correct, even when it may not be.

💸 That is why I think the raw form of “vibe coding” will slowly become less important.

Telling AI “build me something” and waiting for magic is fun for demos and early tests, but it is not enough for serious work.

🛠 Working with AI is becoming a new skill on top of your current skill.

If you are a developer, knowing how to code is no longer enough. You also need to know when to use which model, when to increase effort, when to reduce context, when to limit tools, and when not to let the agent decide by itself.

🚀 I do not think the near future is a world where everyone simply vibe-codes everything.

It looks more like a world where people who already have real experience, and also know how to use AI tools well, become much more effective.

There is a sentence we have all heard many times, but here it really starts to make sense:

AI will not directly take our jobs; but people who know how to use AI well will probably replace people who only kept their old skills.