Tools like ChatGPT and Claude are a subset of AI called Generative AI — more specifically, they're Large Language Models, or LLMs. They're very good autocomplete engines (simplified on purpose) trained on an enormous amount of text. They predict what a helpful, knowledgeable response would look like, based on patterns in billions of documents, books, and websites.

They don't "think." They don't "know" things the way you know things. But they've "read" so much that they can sound fluent on almost any topic, draft almost any document, and explain almost anything in plain language.

The practical: treat AI like a smart, fast, tireless collaborator who has read everything but needs you to tell it what matters in your specific situation. You bring the judgment. It brings the output.

Note: while LLMs are trained primarily on text, many modern models are multimodal, meaning they also "understand" images, code, and audio. The text-trained framing is still accurate enough for most purposes, but worth knowing.

One important caveat: they get things wrong. Sometimes confidently wrong. This is called a hallucination - when a model generates something false but presents it as fact. A made-up statistic, a citation that doesn't exist, a date that's off. It doesn't happen constantly, but it happens.

Always verify anything you plan to act on or share.

The biggest mistake beginners make is treating AI like a search engine — typing a few keywords and expecting magic. Better prompts get dramatically better results. Here's the framework:

Note: while the above is excellent for text-based work, for highly technical tasks — data analysis, coding, logic-heavy problems — consider adding Chain of Thought instructions. Simply adding "Let's think through this step-by-step" encourages the model to work through its reasoning before giving a final answer, which significantly reduces errors in logic.

Don't start by asking "how can I use AI?" Start by asking: what do I do every week that is repetitive, time-consuming, and mostly text?

For most professionals, that's one of these:

• → Writing status updates, reports, or summaries
• → Drafting emails you've written a hundred times
• → Preparing for meetings or creating agendas
• → Summarizing long documents or meeting notes
• → Explaining something complex to a non-expert audience
• → Analyzing a messy spreadsheet — upload a CSV and ask: "What are the top three root causes appearing in this data?" This moves AI from writing assistant to strategic thinking partner.

Pick one of these. Do it with AI this week. That's your start.

Most people try to learn AI the wrong way: they consume tutorials, watch YouTube videos, and read newsletters — all before they have an actual problem to solve. That's just-in-case learning. It feels productive. It rarely sticks.

Just-in-time learning is different. You have a task. You don't know how to do it with AI. So you figure out exactly what you need — nothing more. You use it. It works. It sticks.

Example: "I need to write a performance review for someone on my team. Let me figure out how to prompt AI to help me do that." That's it. That's the whole assignment. You'll learn more in 20 minutes of doing than in 5 hours of watching.

The best way to build confidence with AI isn't to master it — it's to use it on something real and see it work. Even if the first output needs editing. Even if you're not sure you did it "right."

Start with something low-stakes: use AI to draft an email you'd have written anyway. Summarize a document you needed to read. Generate a first draft of a slide deck you'd have stared at for an hour. Then ship it — edited, improved, with your judgment added on top.

That's the loop: Start → Use AI → Edit with your judgment → Ship → Notice what worked.

Do that five times and you'll have a genuine feel for where AI accelerates your work and where it falls flat. There's no shortcut to that knowledge — it just comes from using it.

The most common reason professionals hesitate to use AI at work isn't skepticism — it's a reasonable question: "Is my data training the model?"

The answer depends on which account you're using:

When in doubt: don't paste anything into a consumer tool that you wouldn't put in an email to a stranger. Names, financials, unreleased product details — keep those out of personal-plan tools. Use your company's provisioned account for anything sensitive.

The right mental model for working with AI: AI drafts, you verify. It does the bulk work — generating, summarizing, structuring, calculating. You bring the judgment, the context, and the accountability for the final output.

Think of it like managing a brilliant intern. They're fast, widely-read, eager to help, and can produce impressive first drafts. But they don't know your organization, your stakeholders, or what "good" looks like in your specific context. You wouldn't send their first draft to the CEO without reading it. Same rule applies here.

That said, the workflow isn't one-size-fits-all. Personally, I often prefer to write my first draft myself — then hand it to AI to edit, tighten, or stress-test. It keeps my voice intact while still getting the speed benefit. Try both directions and see what fits your style.