I did a cool thought experiment recently: I tried to predict the future of one specific aspect of AI — prompting.
If you asked me to define prompting, I’d say it’s the art of communicating effectively with an LLM to get a task done well. This definition has a few key parts: there’s a task, a need to communicate it, and a standard for getting it done well.
Today, this is both simple and surprisingly difficult. There are levels to this skill. If you can’t describe a task clearly, even a super-intelligence might fail to execute it. Conversely, with excellent communication, even a system with less raw intelligence can achieve great results. In other words, good communication amplifies intelligence.
This isn’t a new phenomenon. We’ve been using communication to amplify intelligence in our society for ages. When a child struggles with multiplication, we reframe it as repeated addition—a concept they already know. When we introduce new ideas, we lean on analogies because they are easy to grasp and communicate. A perfect modern example is the startup tagline trend, “Cursor for X”. The name “Cursor” has become a shared analogy for an AI-powered tool that boosts productivity.
This kind of communication accounts for the majority of the data LLMs are trained on, so naturally, they operate best on that same distribution. Of course, LLMs aren’t the only conceivable path. One could imagine Large Brain-wave Models that interpret thoughts directly, or Large World Models that predict behavior. This latter approach often models action as a simple chain of events, assuming the next step depends only on the last. But this model is fundamentally limited. It fails to account for true novelty—for ideas that have never been acted upon, concepts that exist only in our minds. To bring those novel ideas into the world, whether with a person or an AI, there is no substitute for effective communication.
This brings us back to the value of prompting today. It’s no secret that “prompt engineering” commands high salaries. You may have seen the in-famous post by OpenAI paying ~$350,000 per annum for “Prompt Engineers”. But this isn’t so different from other fields. Top-tier salespeople, writers, and leaders have always earned a premium for their communication skills. I believe that leveraging LLMs effectively—by communicating with an awareness of their training data to elicit desired results—is a skill that is, and will continue to be, immensely valuable. Looking forward, this dynamic will surely evolve. As the underlying intelligence of these models increases, the nature of communication will shift. It will become less like chanting esoteric magic spells (as some Midjourney prompts feel today) and more like conversing with a hyper-competent specialist—an expert who understands the precise nuance of every word you use.
What do you think? Agree or disagree? I’d love to hear your perspective—feel free to reach out and start a conversation.