AI Is for Everyone
A plain-English guide for people who have never touched it
Two kinds of people are dominating the AI conversation right now, and they are both loud.
The first is the gatekeeper. The tech person, or the consultant, or your family member who works “in computers,” who has decided that AI is far too complicated for you to touch without supervision. Every question gets answered with three more warnings. Every use case comes with a framework, a caution, and a lecture. The subtext is always the same: this is complicated, you are not, call me.
The second is the person at the other end of the bar. The one pasting everything into a chatbot at 11 p.m. with the enthusiasm and judgment of someone six drinks in. Company financials. A customer list. Their kid’s medical records. Straight into a free tool whose terms of service they have never opened. When you ask about privacy, they shrug. Nothing bad has happened yet. They believe AI is their confidant.
I have spent around 30 years in IT and another career in tax, a profession where mishandling data is not embarrassing; it can be a federal crime. (Yes, I know, fear mongering, some say, but I do not write the laws, I just write about them.) So I am allowed to say this: both of these people are wrong. The gatekeeper is wrong because this technology is learnable by anyone who can read, and you do not need permission to learn it. The drunk uploader is wrong for reasons ranging from embarrassing to career-ending, depending on what was pasted.
Between them sits everyone else. The person who hears “AI” forty times a day and could not explain what it actually is if a friend asked. The person nodding along in meetings, hoping nobody calls on them. This piece is for that person. Whatever you do for a living, whatever your age, whatever your comfort with technology. No jargon without a definition. No hype. No doom. No invoice. Just what the technology is, what the big tools are, how the pieces fit together, and what to do with it in your first week.
A Quick Programming Note
Some of you opened this expecting a tax article. Fair. Here is the plan for the next couple of months: I am alternating. An AI topic one week, a tax topic the next. Somewhere along the way the two are going to intersect, because in this profession they already do, and those weeks should be interesting.
This is still Josh & Taxes. The taxes are not going anywhere. But this technology is changing how everyone works, not just tax professionals, and I have spent too many years straddling both worlds to pretend otherwise. If a given week is not for you, forward it to someone it is for. Today’s issue is built for exactly that. Send it to the person who keeps asking you about AI because you are “the tech one,” or to your parents, or to that one coworker. They do not need to know anything about taxes. They just need to be curious.
Now, the actual article.
What AI Actually Is
Start with what it is not. It is not a robot. It is not a search engine. It is not a digital person with opinions and feelings. And despite what half your feed implies, it is not magic.
The AI everyone is talking about right now is a specific technology called a large language model. Here is the plainest explanation I can give you.
Take an enormous amount of text. Books, articles, websites, documentation. Feed it into software that studies the patterns in that text until it becomes extraordinarily good at one task: predicting what words should come next. That is the core mechanism. Prediction.
Your phone already does a primitive version of this. Type “thank you for your” and it suggests “time” or “help.” A large language model is the same idea scaled up by a factor that is hard to comprehend. And at that scale, something surprising happens. A system built to predict the next word turns out to be able to draft a letter, summarize a contract, explain a concept at a fifth-grade level, plan a week of meals around what is already in your refrigerator, and critique your reasoning in a conversational way.
Nobody fully expected that. Capabilities emerged that few predicted, which is why the last few years have felt so sudden. Although maybe it should not have surprised anyone. Humans are very predictable.
So here is the working definition I want you to carry around: modern AI is software that has read more than any human ever could and can generate useful text on demand. It is a very well-read assistant with no judgment, no credentials, and no accountability. That framing is useful, because you already know how to work with a person like that. You give clear instructions. You review the work. You never let them sign anything.
The Big Three
There are dozens of AI tools. For a beginner, three matter. They are more alike than different, and everything in this article applies to all of them.
ChatGPT, made by OpenAI. The one that started the frenzy in late 2022 and the name most people use for the entire category, the way people say Kleenex when they mean tissue.
Gemini, made by Google. Woven into products you already use. If you have noticed AI summaries in Google search results or suggestions appearing in Gmail, you have already met Gemini, whether you wanted to or not.
Claude, made by Anthropic. It is all the rage right now. I use it because it handles nuance well, it writes like an adult, and Anthropic takes safety and data handling generally more seriously than most of the industry. (The bar is low.)
Here is the part that should lower your anxiety: you do not have to pick the right one. All three work the same way. You type, they respond. All three have a free tier. All three have an individual paid plan in the neighborhood of twenty dollars a month. The skills you build in one transfer directly to the others. Choosing between them is like choosing between cars. Real differences exist, people have loyalties, and a beginner should stop agonizing and just start driving one.
The Vocabulary
Eight terms. Learn these, and every AI conversation gets easier to follow. I am sure someone reading this is already typing “well, technically.” Here is the deal. I want people to learn something. I am not here to prove I am smart or that I have all the answers.
AI (artificial intelligence). The umbrella term for software performing tasks we normally associate with human intelligence. Broad enough to be nearly useless in conversation, which is why the next term exists.
Generative AI. AI that creates new content. Text, images, audio. ChatGPT, Gemini, and Claude are generative AI. The fraud-detection system at your bank is AI but not generative.
LLM (large language model). The prediction engine described above. The technology under the hood of all three tools.
Model. A specific version of that underlying brain. Each company makes several, and the version names change constantly. Ignore them. You will pick one from a dropdown menu and mostly never think about it again.
Prompt. Whatever you type into the box. The single most important skill in all of this is writing better prompts. Not coding. Not math. Writing clear instructions, which is a thing you already do every time you ask anyone for anything.
Context window. The tool’s working memory within a single conversation. Everything you have typed, everything it has answered, every file you have uploaded. It is large, but it is not infinite, and it does not automatically carry over when you start a new conversation.
Knowledge cutoff. The model was trained on text up to a certain date and does not inherently know what happened after. Most of these tools can now search the web to fill the gap, but you should always assume an AI’s built-in knowledge has an expiration date. Prices change. Laws change. People change jobs. The model may not know.
Hallucination. The industry’s polite term for the model confidently making things up. A fake statistic. A book that does not exist. A quote nobody ever said. Hallucinations are not rare glitches. They are a known behavior of the technology, built into how prediction works, and they are the single biggest reason the verification habit I describe later is not optional.
The Layers
People get confused about these tools because they conflate four different things: the brain, the meter, the doorway, and the furniture. Every one of the big three stacks the same way.
Layer one: the model. The brain itself. Each company offers a fast, capable default model and a heavier, more powerful one for complex work. As a beginner, use whatever is selected when you sign up. Model shopping is a hobby for later.
Layer two: the plan. The meter. How much you can use the tool, and which models you can reach. Free plans are real, working versions with daily usage limits, not crippled demos. The roughly twenty-dollar monthly plans buy far more usage and the stronger models. Start free. When you hit the usage wall and find yourself annoyed, that annoyance is your signal to consider paying. You will hit it. We all do.
Layer three: the app. The doorway. The same tool is reachable through a website, a desktop app, and mobile apps. There are also developer doorways, such as an API (application programming interface) that lets programmers build the AI into their own software. When any product tells you it is “powered by AI,” it usually means the company built on one of these doorways. You do not need that layer. Start with the website.
Layer four: the features. The furniture inside. Conversations, file uploads, and workspaces, covered next. Features vary by tool and plan and change monthly. Do not try to keep up. Learn three.
Keep the layers separate in your head and the landscape stops being confusing. When someone says “we built it on the API,” that is layer three. When they complain about hitting limits, that is layer two. When they rave about some new capability, that is usually layer four.
Three Features Worth Knowing in Week One
All three tools have some version of each of these.
Conversations. Every exchange happens in a chat, and the tool remembers everything within that chat. So you can refine as you go. “Make it shorter.” “Less formal.” “Now turn that into a checklist.” You do not have to get the prompt perfect on the first try. Treat it like a conversation, not a vending machine.
File uploads. Drag a PDF, a spreadsheet, or a photo into the chat and ask questions about it. Summarize this lease. Compare these two insurance quotes. What does this letter from the school actually want from me? This is where the tool stops being a toy.
Workspaces. Claude calls them Projects. ChatGPT and Gemini have their own versions of the same idea: a folder where you store reference material and standing instructions the AI uses in every conversation inside it. A teacher might load a curriculum. A small business owner might load their price list and brand voice. A family might load the details of an upcoming trip. Every conversation in that workspace starts with the context already in place. This is the feature most beginners find last and wish they had found first.
One housekeeping item before you upload anything: open the privacy settings and read them. Every one of these tools has controls governing what happens to your data, including whether your conversations can be used to improve the models. Understand them and set them deliberately. Two minutes now. Do not skip it.
What to Actually Use It For
You do not need a fancy use case to justify these tools. Start with the unglamorous stuff.
The writing you dread. The email you have rewritten four times because the tone keeps coming out wrong. The complaint letter to the airline. The awkward message to a neighbor. Give the AI the situation, tell it the tone you want, and let it produce a draft. You will edit it. You will still save twenty minutes.
Making long things short. A 40-page contract. A rambling email thread. The 26-page benefits enrollment guide from HR. Upload it, ask for a one-page summary in plain English with the deadlines pulled out, and get on with your day.
Explaining things at your level. This might be the most underrated use of the entire technology. Paste in the confusing thing, the explanation-of-benefits form, the paragraph of legalese, the error message, and say “explain this to me like I have never seen one before.” Then ask follow-up questions. It will not sigh. It will not judge. It has infinite patience for questions you might feel silly asking a person. This is one of my favorite use cases.
Thinking out loud. Stuck on a decision. Unsure how to handle a difficult conversation. Weighing a job offer. Talk it through and ask the AI to argue the other side. It will not make the decision for you. It will make sure you have considered angles you missed at 11 p.m. Please do not mistake this for a therapist. AI can exhibit sycophantic behavior. (Meaning it often wants to agree with you and flatter you excessively.)
The everyday stuff. Meal plans built around what is actually in your pantry. A packing list for a trip with a toddler. A workout plan that accounts for your bad knee. (Not medical advice.) Interview practice before the big one. A birthday toast when you are staring at a blank page. None of this is glamorous. All of it is time back.
And if you have a profession, you have professional uses. I am a tax professional. I use these tools to draft client communications, turn messy processes into written procedures, and translate technical concepts into plain English. Whatever your version of that is, drafting lesson plans, writing listing descriptions, summarizing case notes, it exists, and it probably works better than you expect.
Three Real Examples
Categories are nice. Seeing the actual words is better. Here are three scenarios, with the exact prompt you would type. Notice that none of them require technical skill. They require the same thing good delegation requires: the situation, the audience, and what you want back.
Example one: the airline owes you. Your flight got canceled, the rebooking added a day to your trip, and the airline’s app offered you a coupon. You are angry, and angry first drafts get deleted by customer service. So you type:
“I need a complaint email to an airline. My flight was canceled four hours before departure, the rebooking got me home a full day late, and I paid $180 out of pocket for a hotel. I want a refund of the hotel cost and compensation for the delay. Tone: firm and factual, not angry. Keep it under 200 words. Include a specific request and a deadline for response.”
What comes back is a clean, businesslike letter with your facts organized and your ask stated plainly. You drop in your confirmation number, adjust one sentence, and send it. Ten minutes, start to finish, and the version that went out is the version that gets read instead of flagged. Now, if you have done much traveling, you know this is more about venting and less about getting an actual result.
Example two: the letter that reads like it was written to confuse you. Your health insurer sends an explanation of benefits. It says “this is not a bill” and then lists $2,300 in charges. You take a photo of it, upload it to the chat, and type:
“This is an explanation of benefits from my health insurance. Explain in plain English what happened, what the insurance paid, what I might actually owe, and what questions I should ask if I call them. Assume I know nothing about insurance.”
The tool walks through the document line by line. Allowed amount, what that means. Deductible applied, what that means. Then it gives you three questions to ask on the phone. You did not become an insurance expert. You became a person who can make that phone call without dreading it. Maybe it will help you enough so that you do not have to make that call.
Example three: the blank page at 9 p.m. Your manager asked you to “say a few words” tomorrow at a lunch marking a coworker’s 20 years with the company. You have been staring at your phone for twenty minutes. You type:
“Help me write a short speech for a work lunch honoring a coworker’s 20th anniversary with the company. Two minutes max. Details: she started at the front desk and now runs the entire operations team, she trained half the people in the room, and she is famous for never missing a deadline and never taking credit. Tone: warm and a little funny, nothing sappy. Give me two versions.”
Two drafts come back. Neither is perfect. But the second one has a structure you like, so you reply “keep version two, but end on the people she trained, not the deadlines.” Thirty seconds later you have your speech. You will still deliver it in your own words. The blank page is gone, and the blank page was the whole problem.
That is the pattern in all three: situation, audience, what you want back. No magic words. No secret techniques. If you can brief a person, you can prompt a machine.
The Guardrails, Briefly
This is where I keep you from becoming the person at the end of the bar.
Do not put confidential information into an AI tool without understanding where it goes. Your company’s financials. Customer data. Other people’s medical or legal details. Anything you would not want forwarded outside the room. If your profession has confidentiality rules, and if you are a doctor, lawyer, therapist, financial professional, or tax professional, it absolutely does, then those rules apply to chatbots just as much as they apply to email. In my field, mishandling client data is a federal crime. Yours may be similar. “Nothing bad has happened yet” is not evidence that you are doing it right. It is evidence that you have not checked. I have said this in person to rooms full of people, and I will keep saying it. I am not trying to scare you. I am trying to keep you off the cautionary-tales slide in someone’s presentation.
And remember the vocabulary word from earlier. Hallucination. If a fact, a statistic, a citation, or a quote is going anywhere that matters, verify it against a real source first. Every time. No exceptions. Every tool in this article will eventually hand you something that does not exist, and it will do so with total confidence. The AI is the well-read assistant. You are still the one who signs.
Your First Week
Here is the assignment. Pick one of the three tools. It genuinely does not matter which. Create a free account. Then pick one piece of writing you have been dreading. Give the AI the situation, the audience, and the tone. See what comes back. Refine it twice. Send the version you edited.
That is it. Not a life overhaul. Not a course. One tool, one task.
The people who are getting real value from this did not start with a grand plan. They started with one small task, got a result, and got curious. Curiosity compounds. So does avoidance.
You do not need the gatekeeper. You do not want to be the drunk uploader. You also do not need to sign up for my AI masterclass. (There isn’t one…. yet?) The middle path is not exciting, which is why nobody is loud about it: learn the basics, keep private information out of it, verify everything, and put the tool to work.
You survived learning email. You survived the smartphone. You will survive this. And this one, unlike the others, might actually be fun.
Oh yes. A few things:
If you haven’t checked it out yet, be sure to check out AskTomG.AI. This is one of those projects I wake up excited to work on. We have spent a ton of time and brainpower on this, and it will only get better.
I will be speaking at NATP Taxposium next week! Hope to see some of you there.
Stay tuned for some cool announcements in August.