Getting the Most Out of AI in 2025

There are several distinct types of AI - when we use the words “generative AI” or “gen AI”, we are usually referring to Large Language Models (LLMs) like ChatGPT, Grok, Gemini, or Claude.

If you need more details rather than analogy, you may want to read my more detailed article on how LLMs work .

The simplest way to imagine the current generation of LLMs (Large Language Models) is like a creative, conversational chef that is crafting a dish based on the recipe you’ve asked it to make.

Imagine an LLM as a brilliant chef in a magical kitchen, cooking up responses to your requests. When you give the chef a prompt, it is like you are ordering a specific plate based on an idea. The prompt request starts with a chatbot conversation, typing some words, and handing them a recipe idea, such as “make me a spicy Italian pasta.” The chef then draws from a vast pantry of recipe books and ingredients (their training data) to create something resembling your request.

The chef works by preparing the dish, one ingredient at a time, in a specific order. This is the key to how LLMs generate text.

If you want to sound fancy, there was a seminal paper published in 2017 called “Attention is All You Need”¹ that introduced the transformer architecture. This is the heart of how modern LLMs work. But we digress…

These ingredients are tokens, which are small pieces like words, parts of words, punctuation, numbers, etc. The chef picks each token carefully, based on patterns they’ve learned from countless recipes (their training). They predict what ingredient comes next to make the dish (your response) delicious and coherent.

For example, if you say, “Tell me about leadership” the chef starts with a token like “Leadership” and then predicts the next one, such as “is” ⇒ “about” ⇒ and so on, building the response step by step. The LLM chef’s goal is to serve a complete response that makes sense.

The quality of the dish depends on how clear and specific your initial order is. A vague order like “make something tasty” might get you a decent dish, but a detailed request like “make a creamy pasta with garlic and a hint of chili” helps the chef to get a more accurate response. This is where prompting comes in, because better prompts are like clearer recipes, guiding the chef to produce exactly what you want.

Theory of mind helps a lot here: think about what an LLM agent may know about your business and what they will not know. Tailor your prompts accordingly so that you give the LLM agent enough information to give you a valuable result.

I often think about what the LLM agent cannot know about my use case and I emphasise those points in my prompts. Sometimes I’ve spent 20 minutes+ writing a very long, detailed prompt and that gets me phenomenal results.

To drive the point home, the AI chef will know about general concepts in business, such as what sales people do or what marketing is, however they won’t know how your business organises these functions, or what their unique constraints are, such as territory alignment or regulatory requirements.

The tag #ai-for-busy-people is a series of articles designed to guide business executives through a learning journey about AI, Large Language Models (LLMs), and prompting.

My aim here is to empower you to understand AI and apply it effectively in your businesses.