“Make me a sandwich” : Working with AI Tools

With the rise of AI tools and AI augmentation in the workplace, becoming familiar with how to interact with these technologies is rapidly evolving from a recommended skill to a required one. At the core of most AI tools is the ability to interpret and process natural language—something that can lead users to make assumptions about the AI's capabilities. One of the most common pitfalls is assuming that the AI knows information you haven't provided, which can lead to surprising, and often incorrect, results.

Let's start with the term “artificial intelligence.” Sure, by the textbook definition, the AI is intelligent—it can process data and respond in ways that seem, at times, quite clever. But it's crucial to remember that AI isn't some omniscient oracle. It's a tool, capable of great precision, but only when wielded correctly. And like any tool, its effectiveness is limited by the inputs you give it. If you want the AI to help you, you'll need to spell out the details—literally.

This brings us to a lesson from my youth that has oddly become a metaphor for working with AI tools. In third grade, our teacher gave us a seemingly simple task: "Write down instructions for making a peanut butter and jelly sandwich." After collecting our instructions, the teacher set out a loaf of bread, a jar of peanut butter, and a jar of jelly, then started following the instructions exactly as written.

Some students wrote, “Put peanut butter on the bread.” So, the teacher placed the unopened jar of peanut butter on top of the wrapped loaf of bread and moved on. Others said, “Spread the peanut butter,” without specifying how to open the jar or what to spread it with. The results were comedic, but the lesson was clear: You can't rely on assumptions or unstated knowledge, and instructions must be explicit.

Fast-forward to today, and we find ourselves writing instructions for AI systems that are just as literal-minded as my third-grade teacher. AI doesn't interpret, it doesn't infer, and it most certainly doesn't assume. It processes. So, when you ask an AI to "write a thank-you note to Tom," and it responds with “Dear Tom, thanks,” don't blame the AI for its lack of finesse. You didn't tell it who Tom is, what he did, or why you're thanking him; provide all the relevant data. For all the AI knows, Tom could be your mechanic, your CEO, or a goldfish you mistakenly anthropomorphized.

Now, let's apply this to a more technical scenario: interacting with AI tools in a professional environment. Much like programming, using AI effectively requires precision. The lesson from that peanut butter and jelly experiment is directly applicable: define everything. Assume nothing. If you want a poster for your Halloween party, don't say, “Make a spooky Halloween poster.” That's asking for trouble. Instead, be explicit: “Create a poster with a night-time forest scene using blue and green tones, a full moon in the sky, and the text ‘Halloween Party' in Ghoulish Fright font over the moon.” Be so clear that even a robot could follow it.

In programming, this concept is fundamental—precise instructions lead to predictable outcomes. Ask any programmer who's spent hours debugging a line of code that's doing exactly what they told it to do, but not what they wanted it to do. Writing for AI is a similar exercise in humility. You might feel that you've given the AI enough information, but when it gives you something entirely different from what you envisioned, it's probably because your instructions were too vague or ambiguous.

Here's a pro tip: Allow for iteration. No AI tool—no matter how sophisticated—is likely to get it perfect on the first pass, just like most developers don't write flawless code the first time around. If your output isn't what you expected, tweak your input. Think of working with AI as a cycle of refinement. You wouldn't submit your first draft of a report to the CEO without a revision, and you shouldn't expect AI tools to hit a home run on the first try either.

To sum up, AI tools are powerful, but they are not magical. While they can make associations between information in order to make predictions on what comes next, they cannot know something they're not explicitly told about, such as knowing that peanut butter belongs inside the sandwich. The key to working effectively with AI is clear, detailed instructions, information about the subject the AI model is working on, and enough context for the information provided to allow the AI model to find those patterns and associations. Don't assume the AI shares your context.

AI won't make your sandwich unless you spell it out, and while it may never appreciate the fine art of lunchtime assembly, it can help you with plenty of other tasks—provided you tell it what, exactly, you need. Give it enough information, and it might just help you write that thank-you note or design that poster while you actually make a sandwich.