Meet your AI staff: your first real conversation

You've probably asked ChatGPT a question before — a recipe, a settled argument, a quick fact. Today you do something different: you set up all three major AI assistants for free, get an honest explanation of what's actually inside them, and have your first real working conversation — the kind where the tool asks you questions and hands you a draft you can actually use.

1Search vs. delegate: two different jobs that happen to share a text box

Most people meet AI assistants through a search-shaped door. The screen looks like Google — a box, a button — so they use it like Google: type a few words, skim whatever comes back, close the tab. If that's been you, nothing was wrong with it. It's the natural first move. It's also a tiny fraction of what you've been sitting on.

Here's the distinction the whole course is built on. A search engine retrieves documents that already exist — pages written by other people, for a general audience, before you ever showed up. The reading, the comparing, the deciding, the writing? All still yours. An AI assistant does something else entirely: it produces new work, shaped to your situation — a draft of your email, a plan for your week, an explanation pitched at your level, a list of options with the trade-offs spelled out. And it can do something no search engine has ever done for you: it can ask you questions before it starts.

One bit of vocabulary before the picture: the message you send to an assistant is called a prompt. Over the next ten days, your prompts will evolve from queries into something closer to briefings. Here's what the two loops look like side by side:

Look at where the effort sits in each column. On the left, the tool hands you raw material and the labor is yours. On the right, you spend your effort on two things only: describing what you need and judging what comes back. That's delegation — and delegation is a skill. Most people never learn it because nobody told them the box wanted full sentences. By tonight, you'll have done it once for real.

2What's actually inside the box — the honest version

You don't need any technical background for this course, but you do deserve one honest paragraph about what you're talking to — because that one paragraph explains both the superpowers and the flaw.

An AI assistant is a system trained on a huge amount of text — books, articles, websites, conversations — that has learned, from all that reading, to produce useful, fluent responses to whatever you write. That's it. It is not a database of verified facts. All three assistants can now reach out to the live web when a question needs fresh facts — you'll sometimes see them do it on their own, with sources attached — but the words you read are still generated, not copied from a verified record (Day 4 covers when to demand a live search and how to check the sources). It is producing the most useful, plausible response it can, based on everything it has absorbed.

That training is why it can do things no other tool you own can do: draft a difficult email in your situation, rewrite a paragraph five ways, explain a pension document in plain language, brainstorm names for your project at midnight without sighing. It has, in a meaningful sense, read almost everything — so it can meet you on almost any topic.

And the same fact explains the flaw. Because the assistant generates responses rather than looking up certified answers, it sometimes produces a statement that is simply false — a date, a price, a quote, a "fact" — delivered in exactly the same fluent, confident tone as everything true it says. There's a technical word for this, and it's the one piece of jargon worth learning today: a hallucination — when an assistant states something false with total confidence. For the rest of this course we'll mostly call it what it feels like: a confident mistake.

Put the two halves together and you get the master frame for these ten days — the Confident Intern. Hold that frame and three consequences follow, each owning a chunk of this course:

• You brief an intern — you don't keyword-search them. → Day 2 gives you the five-ingredient Briefing Formula.
• You check an intern's work before it leaves your desk. → Day 4 gives you the Three-Click Check, a complete, fast verification habit.
• An intern gets better the longer they work with you. → Day 9 sets up memory and workspaces so yours does too.

Not magic. Staff. That mental shift — from "mysterious oracle" to "capable colleague whose work I direct and review" — is the single most valuable thing you can take from today.

3Meet the big three — your whole staff, for $0

Three assistants matter for this course, one from each of the major AI companies:

• ChatGPT, by OpenAI — the most widely used assistant in the world and a strong all-rounder.
• Claude, by Anthropic — many people's favorite for writing, careful analysis and working with documents.
• Gemini, by Google — the one woven into the Google world, with notably generous free extras.

Here's the part that surprises people: all three have free plans, and this entire course runs on them. Every exercise in the next ten days costs $0. The free plans do have usage limits — a small allowance of messages or uploads that resets on a timer — and that allowance is normally plenty for everything we'll do. The exact limits change often, so don't memorize a number: check the current limits inside each app if you ever hit one. If the figures look different on your screen, the workflow still works the same.

Why all three, instead of picking one now? Because they're staff, not gadgets — each has genuinely different strengths, and on Day 3 you'll compare them side by side on the same tasks and choose your daily driver: the one that becomes home base. Today you set up all three accounts so that choice is yours to make with your own hands, not someone else's review.

One promise about how this course talks about these tools. The companies behind them release new versions constantly — new model names appear every few weeks, like a software changelog that never stops scrolling. So this course will almost never mention a version number: any name we printed would be stale by the time you read it, and — more importantly — nothing about your skills changes when the version does. We teach capabilities and habits, not version numbers. On Day 8 you'll hear the story of a famous AI video tool that went from "the future of video" to discontinued within a year — and why the people who'd learned workflows rather than apps didn't lose a thing.

4The thread remembers — and iteration is the superpower

One more concept and you're ready for the keyboard. A single conversation with an assistant is called a thread (or a chat). Within one thread, the assistant remembers everything that's been said — your situation, its earlier answers, the corrections you made. A new chat starts mostly fresh — though some assistants now carry a light memory of your past chats by default, so don't be surprised if one references something you said yesterday. (Day 9 shows you where that memory lives and how to control it; today, one thread is all you need.)

Because the thread remembers, you can do things that feel almost rude the first time, and that's exactly the point:

• Ask follow-ups without re-explaining anything: "and what about the cost side?"
• Push back: "you've assumed I'm the landlord — I'm the tenant."
• Ask for a redo: "make it half as long," "warmer," "give me three versions."
• Change direction mid-stream: "actually, let's turn this into a phone script instead."

This is iteration — replying with feedback to improve the answer instead of starting over — and it is the superpower of Day 1. The single biggest difference between people who get a little out of these tools and people who get a lot is not the first message they type. It's what they do after the first answer. The first answer is a first draft. Always. Redos cost nothing, arrive in seconds, and are never met with a sigh — and unlike a human colleague, the assistant never gets defensive about feedback at 11pm.

If today leaves you with exactly one habit, make it this: reply. The send button isn't a verdict on the tool or on you — it's one turn in a conversation that gets better the longer you stay in it.

Part 1 — Create your three free accounts (~5 minutes, one-time)

You'll do the same short dance three times. Each account is free, and each signup ends in the same place: an empty chat window.

1. In your browser, go to chatgpt.com. Find the sign-up option and create a free account — you can use an email address and a password, or the "continue with Google/Apple" shortcut if you prefer. Confirm your email if it asks. a chat window: an empty message box near the bottom of the screen, and a sidebar (or menu) on the left where your past chats will live.
2. Open a new browser tab and go to claude.ai. Same dance: sign up free with an email or a Google account. the same shape — a message box waiting at the bottom, a sidebar for chat history. Different colors, same anatomy.
3. One more tab: gemini.google.com. If you're already signed in to a Google account (Gmail counts), you may land straight in the chat; otherwise sign in or create a Google account. the third chat window of the day. You now have a staff of three, and you've spent $0.
4. Bookmark all three tabs, or pin them. You'll visit all of them this week. three bookmarks — chatgpt.com, claude.ai, gemini.google.com — sitting in your browser like three new hires waiting for instructions.

Part 2 — The anatomy of a chat window (the same in all three)

Before you type anything, take thirty seconds to find seven controls. Every chat assistant has all seven; they shuffle position every few months as the apps redesign, so learn the icon and the idea, not the pixel location.

Part 3 — The first real conversation (in ChatGPT)

We'll run this first one in ChatGPT, then wave at the other two. Before you start, pick one real task or decision from your actual week. Not a test question — something genuinely on your plate: an email you owe someone, a tricky message to a relative, a purchase you keep researching, a small event you're supposed to organize. Real material is the fuel of this whole course.

1. Go to your ChatGPT tab and click the new-chat button so you're starting clean. Click into the message box, then type (or copy, paste and fill in) this first prompt:
   Copy-paste prompt · ChatGPTI want your help with something real. Here's the situation: [describe one real task or decision from your week in 2–3 sentences]. Before you start, ask me up to 3 questions that would help you help me.
   not an answer — questions. Two or three of them, specifically about your situation. Sit with that for a second: a search engine has never once asked you a question back.
2. Answer its questions the way you'd answer a colleague — plain sentences, no special formatting, typos welcome. If a question doesn't apply, say so ("doesn't matter in my case"). it absorbs your answers and responds — often summarizing your situation back to you, correctly, in cleaner words than you used.
3. Now ask for the actual work:
   Copy-paste prompt · ChatGPTGreat. Now give me a short plan and a first draft I can actually use.
   a brief plan plus a usable first draft — built around the details you gave it, not a generic template with your topic sprinkled on top. This took seconds. If you see little source links, that's the assistant deciding to check the web — more on that on Day 4.
4. Read the draft like an editor, not like an audience. Then put iteration on display:
   Copy-paste prompt · ChatGPTMake it half as long and friendlier.
   a fresh version, instantly, with no complaint — half the length, warmer tone. Redos are free. This is the superpower from section 4, live on your screen.
5. Optional but recommended: push back once in your own words — correct something it assumed, or ask for one more variation ("give me a version I could text instead of email"). the draft bending to your feedback again. You're not searching anymore. You're directing.

Part 4 — Rename the chat so you can find it again

1. Look in the sidebar: your conversation is there, with an automatic title the assistant invented.
2. Open the little menu next to the chat title (usually a "⋯" or a right-click on the title) and choose Rename. Give it a name future-you will recognize — "Landlord email" beats "Conversation 1". your own words at the top of the sidebar. That list is about to become an asset: threads you return to, not transcripts you lose.

And the other two? Claude and Gemini work exactly the same way — same message box, same conversation loop, and the prompts you used above would work in either, word for word. If you're curious tonight, paste the first prompt into one of them and watch a different personality handle the same brief. On Day 3 we'll do that comparison properly, with a scorecard.