AI saves you effort—and sometimes robs you of a skill

AI can complete a task faster than a human, but that doesn’t necessarily mean it’s worth assigning it to AI. Sometimes the result itself is what matters, and sometimes the value lies in a person becoming capable of producing that result.

The article proposes a useful distinction: work or training.

If a task is like work—where getting the finished result is what matters and the method doesn’t—automation makes sense. Instructions, presentations, formal documents, or other predictable texts can be entrusted to AI if it’s reliable enough and errors can be checked.

If the task is more like training, automation misses the mark. A student writes an essay not because the world urgently needs yet another text. They’re learning to articulate their thoughts: to build an argument, identify weaknesses, revise, and verify their own conclusions. By handing this part of the process over to a chatbot, you might end up with polished text, but you’ll miss out on the learning process itself.

This explains why a confident and grammatically flawless response can be of little help. It eliminates the uncomfortable phase in which a person realizes they don’t yet understand the topic or can’t clearly express their thoughts. But it is precisely this phase that often constitutes the work of thinking.

Before using AI, it’s helpful to go through a quick screening process:

• Do you only need the final output?
• Or is the skill of completing the task also a goal?
• Can you independently verify the result and correct any errors?
• Without practice, will the ability to do this on your own disappear?

This approach has an unfortunate economic consequence. In the past, people were often hired to work on text and images regardless of whether the task required genuine creativity or simply careful execution. AI distinguishes between these cases: it can take on routine work, while human value will increasingly depend on creativity, judgment, and a willingness to keep practicing.

Therefore, “don’t use AI” is too blunt a rule. It makes more sense to let AI handle the work while you do the training yourself. In development, this means not shunning tools, but understanding which tasks test and develop engineering thinking, and which merely move data from one place to another.

The main limitation is obvious: all of this only makes sense if the AI actually handles the task, its errors are minimal, and the system is protected from influences that alter the result. But even a reliable model doesn’t answer the question that’s often more important than the technical one: what exactly are we trying to achieve right now—a finished product or the ability to do it better?

Source: Hacker News - Newest: ""AI" "LLM""