All lessons

What AI is (and isn't)

Lesson 1 of 17 · 9 min

Tools like ChatGPT, Microsoft Copilot, Google Gemini and Claude are built on large language models. They have learned patterns from an enormous amount of text and generate responses by predicting what words should come next. That single idea explains why they are so useful and why they sometimes go wrong.

Around that core, today’s assistants have picked up extra abilities. Many can search the web, read files you give them, remember your preferences and, in some tools, take actions such as filling in a form or editing a document. These additions are genuinely helpful, and each one gives you something new to check.

What has changed, and what has not

  • Changed: they can look things up. Many assistants search the web and show their sources. That helps, but the answer is still written by a model that can misread a page.
  • Changed: they can use your material. You can attach documents, spreadsheets and images, and, if you allow it, connect email, calendar and files.
  • Changed: they can remember. Some tools keep notes about you across conversations.
  • Changed: some of them act. Agent features can work through several steps for you. Lesson 11 covers how to use them safely.
  • Changed: answers can be interactive. Charts, tables, calculators and small tools can appear inside a reply, not just paragraphs.
  • Not changed: they can still be confidently wrong, and you are still the one accountable for the result.

What they are great at

  • Drafting. First versions of emails, reports, job descriptions and social posts.
  • Rewording. Making text shorter, clearer, friendlier or more formal.
  • Summarizing. Turning long threads, documents and recordings into the key points.
  • Brainstorming. Generating options, names, angles and questions you had not thought of.
  • Explaining. Breaking down an unfamiliar topic at whatever level you need.
  • Structuring. Outlines, checklists, agendas and plans.
  • Spreadsheet help. Writing and explaining formulas, and spotting problems.
  • Reading what you show it. A PDF, a chart, a screenshot or a photo of a whiteboard.

Where they fall short

  • They can be confidently wrong. A model writes what sounds right. It can invent names, numbers, quotes and sources that look perfectly plausible. This is often called hallucination, and it has become rarer without going away.
  • They can misread what they find. Even with web search, a tool can summarize a page incorrectly or lean on a weak source.
  • They do not know your situation. They cannot see your company, your customers or your history unless you tell them or connect them.
  • They reflect patterns in their training data, including biases and clichés.
  • They can be steered by what they read. Text hidden in a web page or document can try to give a tool instructions. Lesson 4 and lesson 11 explain this in plain terms.
  • They are not people. They have no judgment about consequences and no accountability. You do.

The right mental model

Treat AI like a very fast, very well-read colleague who started this morning. They produce a solid first draft in seconds. They need clear instructions, they do not know how things work around here, and everything they hand you needs your review before it goes anywhere. When a tool can also act, picture that colleague holding the keys to your accounts. You decide how much access they get.

AI does not replace your expertise. It multiplies it. The more you know about your work, the better you can direct it and the faster you can spot when it is wrong.

Myths worth dropping

  • “I need to be technical.” You do not. If you can write an email you can use these tools.
  • “It will take my job next week.” Nobody can promise how work will change. The people gaining an edge right now are the ones who use these tools well and check their output. You are learning to be one of them.
  • “I can’t break it.” True of the tool, with one caveat. You can overshare, or let an agent act too freely. That is why the safety lessons come early.
  • “If I don’t know it perfectly, I’m behind.” Nobody knows it perfectly. The tools change every few months, and most people are still working it out.

Try it

Ask an AI tool to explain your job to a 10-year-old. Notice what it gets right, what is generic, and what is wrong. That gap is where your expertise lives.

Key takeaways

  • AI is excellent at drafting, rewording, summarizing, brainstorming and structuring.
  • Search, memory, connections and agents are add-ons. Each one is useful and each one needs a check.
  • It can sound confident and be wrong, so you always check.
  • You stay accountable for everything you use.

Get the free Safety Checklist

One printable page: what never to paste in and how to check an answer.

We'll only email you about this. Unsubscribe anytime.