When we talk about prompt engineering, at its core we are talking about the art of bringing machines to life through language. But is it really art, or is it more about sober, technical craftsmanship? As is often the case, the truth lies somewhere in between.

The creative side: Prompting as an art form

A good prompt can sometimes feel like a poem. It lives through nuance, metaphor, and the right tone. Anyone who has worked with prompts before will have noticed this. It is not just about retrieving information, but about setting a mood, giving direction, and nudging the machine into a specific way of thinking.

Experimentation is often an essential part of the process. Translating your thoughts or your vague idea of what you want into words that a model truly understands, and in a way almost feels, is not easy. Sometimes you do not even know exactly what you want at the beginning. The final result often emerges from the flow of the conversation and the inspiration it generates.

A prompt can act like a director’s instruction. Each individual cue guides the model through your mental image and, ideally, produces a detailed representation of what you had in mind.

There are no fixed rules or formulas that guarantee a specific outcome. That is precisely what makes prompt engineering fascinating for many, but also difficult to grasp and sometimes frustrating. It gives the whole topic a slightly mystical aura. It is much like painting a picture. There are countless paths to the same destination. These paths may look very different, yet still arrive at similar results. Just as a painter plays with colors, a prompt engineer plays with language, and that requires a certain level of creativity.

The technical side: Prompting as a craft

Even without strict rules, there are proven techniques, just like in painting. These techniques make it possible to reach specific results faster and more reliably. While experimentation is often valuable, in many cases it can be reduced or at least kept under control.

One key concept here is structure. Well structured prompts tend to produce more precise results. Even though we are communicating with a machine, the process is surprisingly similar to human communication. When you present your thoughts in a structured way during a conversation, the other person understands you more easily and needs fewer clarifying questions than if you simply talk without order and overwhelm them with scattered ideas. Even if the amount of information is the same, extracting meaning is much harder in the second case. This applies to humans, and it applies to AI as well.

Clarity is just as important as structure. The more precise the instruction, the more consistent the output. Vague statements always leave room for interpretation, and that is exactly what the model will do. It will interpret, and that interpretation may differ from your own. When you clearly state what you want, you reduce ambiguity and increase the likelihood that the model delivers exactly what you had in mind.

This becomes especially obvious in image generation or other visual media. A prompt like “Create an image of a tree” will indeed result in a tree of some kind, but what kind of tree it is, where it stands, what the background looks like, and which style is used is entirely open to interpretation. The model’s mental image may differ greatly from yours. A prompt such as “Create a digital painting of a cherry tree in full bloom, in a Japanese garden style, with a soft sunset in the background” already narrows things down significantly. The chance of getting a palm tree on a beach is rather low. Whether there is a bird in the image or not, however, is still undecided.

This shows that the more detail you provide, the more precise the result becomes. Beyond these general best practices, and there are many more, there are also prompting strategies and frameworks that shape prompts on a meta level. This is where things can get truly technical. These approaches are often what separate a good prompt from a very good or even excellent one.

By the way, what actually defines a good or excellent prompt? Are there metrics for that, and how can something be evaluated objectively when the output is ultimately a reflection of one’s own thoughts? These questions will be explored in future articles, so stay tuned.

Methods such as chain of thought, few shot prompting, or prompt chaining function almost like programming logic. With them, prompt engineers can build systems that solve very specific and sometimes highly complex tasks with meticulous precision. Examples include programming assistants like GitHub Copilot, design tools like Vercel’s V0, or advanced chatbots that guide customers autonomously through ordering processes or onboarding flows.

For such applications, repeatability and predictability are essential. AI model outputs are not deterministic by default. The same input can lead to different results. In some scenarios, this variability is undesirable, especially when decisions are made based on the output or when responses must follow a specific format. By combining multiple techniques, it is possible to approximate a high degree of consistency.

The intersection: Where art meets engineering

This is what makes prompt engineering so intriguing. It sits precisely between these two worlds. It is neither pure art nor pure engineering, but a hybrid of both.

Prompt engineers are creative technologists. They are designers with technical expertise. On one hand, they can translate complex thoughts or abstract ideas into written instructions that are so clear, precise, and unambiguous that the AI understands them and produces exactly the intended result, much like a programming language. On the other hand, they can create worlds and ideas that were previously unimaginable by playing with both their own creativity and that of the AI, much like an artist.

It is a constant balancing act. Those who think only in technical terms miss out on creative magic. Those who focus solely on artistic expression lose precision. True strength lies in combining both.

“The hottest new programming language is English.”
Andrej Karpathy