The other half of the job

Chloe'sguide to AI

Not what it is. What to do with it, and what stays true once today’s tools look old.

Ask, or show your working

Ask anything about using AI. Or switch to Critique my prompt, paste in something you wrote, and it will tell you why the answer disappointed you and rewrite it properly. That second one is the fastest way to get good at this.

It can be wrong, like anything, and it is the first to tell you so. If an answer contradicts the guide below, trust the guide. Nothing you type here is stored anywhere.

01Start Here

Chloe

You said the camera guide was awesome and that I should be selling it.

Here's the part I didn't tell you. It took an evening. Not an evening of typing, an evening of deciding what it should be while something else did the typing. The 10,000 words, the diagrams, the thing that reads your photos and tells you what went wrong, the website itself. One evening.

That's the whole point of this second guide.

I'm not going to tell you AI is the future or that it'll change everything. You've heard that from people trying to sell you a course. What I'll tell you instead is narrower and more useful: there's now a gap between what these tools can do and what most people know how to ask them for, and that gap is where the money and the interesting work sit. Not for long, probably. But right now, yes.

Your degree isn't wasted here. It's the opposite. The world is full of people who can make a computer do things and can't tell you why a business would want it. You're being trained in exactly the half that's scarce. Add this to it and you have something almost nobody has.

How this is different from the camera guide

The camera guide was facts. Where the buttons are, what f/1.8 does, what happens at 1/30. Those don't change. I checked them against Canon's manual and they'll still be true in ten years.

This one rots. The specific tools will be different in a year and unrecognisable in three. So I've written it in two halves:

If you only read one part, read the asking. It's the actual skill.

The box at the top

Same as the camera guide, with one addition that matters more here.

Paste in a prompt you wrote and it'll critique it. Tell you why it gave you something mediocre, what you left out, and rewrite it properly.

Use that a lot. It's the fastest way to get good at this, because you're learning the skill by doing the thing, with someone marking your work each time. I had nothing like it and it took me months longer than it needed to.

Grandad

02What This Thing Actually Is

Before the how, a bit of the what. Not the technical version, the useful version. Most people's mental model is wrong in a way that makes them worse at using it, and fixing that is worth more than any list of tips.

It is not a search engine and it is not a person

Two wrong models, and nearly everyone holds one of them.

"It's a clever search engine." If you think that, you ask short questions and accept the first answer, the way you would with Google. You get shallow results and conclude it's overrated.

"It's a person who knows things." If you think that, you trust it the way you'd trust a knowledgeable friend, and it will eventually tell you something wrong with total confidence and you'll believe it.

Neither is right. The closest useful model:

It's an extremely well-read, very fast collaborator who has never met you, doesn't know what you're working on, has no memory of yesterday, and will never admit to being tired.

Every part of that sentence changes how you should use it.

Well-read means it genuinely knows things. Ask it about Portuguese trade law or how a shutter is manufactured and it'll know more than you expect.

Very fast means the cost of asking is nearly zero. Ask five ways. Throw four away. That's not wasteful, it's the correct technique, and it's the habit most beginners never form.

Never met you means it has no idea what you're doing unless you say. This is the single biggest cause of disappointing answers, and it's covered properly in the next chapter.

No memory of yesterday means every conversation starts from nothing.

Never tired means you can ask it to do the thing again, differently, twelve times, and it will not sigh. Use this. People are astonishingly reluctant to.

What it's genuinely good at

What it's bad at, and this matters more

The mistake nearly everyone makes

They ask for the answer.

The people getting real value ask for the work. Not "what should I write about X" but "here are my notes, my audience, three things I've read, draft this and then tell me what's weak about it".

The difference isn't cleverness. It's how much you put in.

Grandad's version of this

I keep saying it because I believe it and because it's the useful frame: we've built these things in our own image, and the human is now the throttle, not the engine.

The capability is sitting there. What limits what gets made is how well a person can say what they want, judge what comes back, and decide what to do next. That's judgement, taste and clarity, and it's all yours.

Which is why this guide spends more time on how to ask than on which tool to use.

03How To Actually Ask

This is the chapter that matters. Everything else is context.

The skill isn't secret and it isn't difficult. It's just that almost nobody does it, because typing one line feels like enough and finding out otherwise takes a few months.

The four things a good ask contains

Most bad answers come from a missing one of these.

1. Who it should be. "You're a photography teacher explaining to a complete beginner." Not magic, but it sets vocabulary, depth and tone in four words.

2. What you actually want. Not the topic, the outcome. "Explain aperture" is a topic. "Explain aperture so I understand which number to turn and why, in about 200 words, no jargon" is an outcome.

3. What it needs to know. Your situation, your constraints, what you've already tried. It has never met you. It cannot infer that you're on a phone, in a hurry, writing for your tutor, or that you already rejected the obvious answer.

4. What good looks like. Length, format, tone. Or better, an example. "Like this, but about X" is the highest-leverage sentence in this entire guide.

You don't need all four every time. You need to notice which one you left out when the answer disappoints.

Before and after

Weak:

How do I make my Instagram better

Strong:

I post travel photos, mostly landscapes and my friends, taken on a Canon compact. About 400 followers, mostly people I know. I want more of the right people finding it, not just numbers. I post maybe twice a week and I'm not willing to do daily stories. Give me five things to change, ordered by effort against payoff, and be honest if something won't work at my size.

The second gets a genuinely useful answer. It isn't cleverer. It just says who, what, the constraints, and what good looks like.

Notice the last clause. "Be honest if something won't work" gives it permission to disagree with you, and without permission it tends to be agreeable rather than useful.

The techniques worth knowing

Ask for the thinking, then the answer. "Work through this step by step before you conclude." Genuinely improves anything with reasoning in it.

Ask for options, not an answer. "Give me three approaches, with the tradeoff for each." Then choose. You'll make a better decision than it would.

Make it argue with itself. "Now give me the strongest case against what you just said." Brilliant for spotting where it's been agreeable rather than right.

Give it an example. One example beats a paragraph of description every time.

Tell it what not to do. "Don't use bullet points." "Don't hedge." "Don't explain what you're about to do, just do it."

Ask it what it needs. "What would you need to know to do this properly?" Then answer that. Underused and unreasonably effective.

Iterate. Never accept the first draft. The first is a starting point. "Now make it half the length" or "the third paragraph is weak, fix it" is where the quality comes from. This is the habit that separates people who find it useful from people who find it disappointing.

The rule you must never break

Anything factual, check.

Especially numbers, dates, prices, names, quotes, statistics, legal or medical claims, or anything you'd be embarrassed to be wrong about in public.

It will produce a plausible figure with total confidence. Not because it's lying, because producing plausible text is what it does, and a plausible number looks exactly like a real one.

The camera guide is trustworthy for exactly one reason: every camera fact in it was checked against Canon's own manual rather than recalled. That's the difference between something useful and something dangerous, and it took an extra hour.

Apply this hardest when it tells you what you were hoping to hear.

Ten minutes that will make you better

Take something you asked recently that gave a mediocre answer. Paste it into the box at the top of this page and ask it to critique your prompt.

Then rewrite and ask again.

Do that five times over a week and you'll be better at this than most people who've been using it for a year.

04The Ladder

Five levels. Most people stop at two and conclude they've seen what it does.

The jump that matters is from three to four, and almost nobody makes it.

Level 1: Asking things

Questions, explanations, quick answers. Useful, and where everyone starts.

You're here if you use it like a better search engine.

To move up: start giving it your actual context instead of generic questions.

Level 2: Making things with it

Drafts, summaries, rewrites, plans, structure. You bring the raw material, it does the shaping.

You're here if it saves you real time on real work.

Most people who say they "use AI" live here permanently. Nothing wrong with it, but it's the foothills.

To move up: stop accepting first drafts. Iterate three or four times on the same thing.

Level 3: Working with it properly

You've stopped treating each question as separate. You give it a role, feed it your material, argue with the output, ask it to attack its own answer. Conversations rather than queries.

You're here if you'd genuinely miss it, and if you notice when a bad answer was your fault.

To move up, and this is the real jump: stop asking it for text. Start asking it for things.

Level 4: Building things

The gap most people never cross, and the reason for this guide.

It doesn't only produce words. It produces working things: websites, spreadsheets that calculate, tools that do a job, scripts that automate something tedious. You don't need to be able to code. You need to be able to describe what you want, look at what came back, and say what's wrong with it.

That's the same skill as level 3, pointed at something that runs.

You're here if you've made something that works, that someone else could use.

Your camera guide is a level 4 artefact. It's a website that reads photographs. I couldn't have written that by hand and I don't need to be able to.

To move up: solve someone else's problem, not your own.

Level 5: Solving other people's problems

The level where it stops being a skill and starts being worth money.

The scarce thing is not the tools. Everyone has the same tools, and yours cost about the same as a coffee. The scarce thing is someone who understands a real problem well enough to know what to build for it.

This is where everything else you're learning pays off. The world has plenty of people who can make things and can't tell you which thing is worth making. Understanding people and what they actually need is the scarce half, and it is not a technical skill.

You're here if someone who isn't related to you has asked you to do it again.

The honest bit about levels 4 and 5

Level 4 is genuinely achievable in weeks, not years. That's not encouragement, it's just true now in a way it wasn't three years ago.

Level 5 is harder, and not for technical reasons. It's harder because it needs you to find someone with a real problem and understand it properly, and that's slow, human work that no tool shortens.

Most people who get stuck aren't stuck on the technology. They're stuck because they don't know anyone with a problem worth solving.

Which means the useful thing to do at level 4 isn't more practice. It's talking to people.

05Build Something This Week

Reading about this does almost nothing. The whole thing clicks the first time you make something that works, and that can be this week.

Start with the smallest real thing

Not a startup. Not an app. Something small, real, and yours.

The test is: would you actually use it, or is it homework? If it's homework you'll abandon it, and rightly.

How to actually do it

Describe the finished thing first, in plain English, before anything else. What is it, who uses it, what does it do. Two or three sentences.

Ask what it needs to know. "What do you need from me to build this properly?" Answer that honestly.

Get the simplest version working before anything else. Ugly and working beats beautiful and theoretical.

Then say what's wrong with it. This is the whole job. "The middle bit is confusing." "This should be one screen not three." You are the taste and the judgement. That's not a lesser role, it's the harder one.

Expect it to take a few goes. The camera guide took several. Diagrams needed redrawing, the answers rendered wrong the first time, the question box was clipping its own text. Normal. The difference between something good and something rough is entirely in the going-back-again.

The thing that will stop you

Not difficulty. Not knowing what to make.

Everyone hits this. The fix is unglamorous: keep a note on your phone, and every time something is annoying or takes too long or you think "why isn't there just a...", write it down. Don't judge it, just collect.

Within a fortnight you'll have a list. Two or three will be buildable in an afternoon.

That list is worth more than any course.

What to do when it goes wrong

It will. Something won't work and you won't know why.

Say exactly what happened. Not "it doesn't work" but "I expected X, I got Y, here's what I did." The same rule as asking well.

Ask it to explain rather than fix. "Why is this happening?" teaches you something. "Fix it" doesn't, and you'll hit the same wall next week.

If you're going round in circles, start fresh. Sometimes a conversation gets tangled. Open a new one and describe the problem cleanly. Works more often than it should.

06Six To Start With

Things worth building, chosen for you rather than for a generic beginner.

1. A trip planner that isn't generic. Feed it your actual dates, budget, what you like and what you can't stand. Get a plan that accounts for you rather than for a tourist. Then keep refining until it's genuinely good. You've got California and Canada coming.

2. A study tool for your course. Take a topic you find slippery. Get it explained three ways, turned into questions, then have it mark your answers and tell you what you've misunderstood. Better than any revision guide because it's built round what you don't get.

3. A thing that does your worst repetitive task. Whatever you do the same way every week. There's almost always something.

4. A packing list that thinks. Sounds trivial, isn't. Destination, season, activities, luggage limit, what you always forget. Now it's a tool, not a note.

5. A one-page website for something you care about. This is the biggest jump for the least effort, and it's the one I'd push you towards. You've seen what one looks like.

6. Something for someone else. A friend's problem, however small. This is the level 4 to 5 jump in miniature, and it's the only one on this list that teaches you something the others can't.

07When It Becomes Worth Money

Understanding people is the useful half here, so this chapter is shorter than you'd expect and blunter than most things written about it.

The uncomfortable truth first

Being good with AI is not a job. It's rapidly becoming like being good with a spreadsheet: expected, unremarkable, and not something anyone pays a premium for on its own.

Anyone telling you otherwise is usually selling a course.

What people pay for is the same thing they always paid for: a problem of theirs, gone. The tools change how fast you can make that happen. They don't change what's being bought.

So the question isn't "how do I monetise AI skills". It's the old question: whose problem do I understand well enough to solve?

Where you actually have an advantage

Three, and they're real.

You notice what actually annoys people. Enormous numbers of technically capable people cannot tell you why anyone would want the thing they've built. Noticing the real problem is the scarce half, and it never feels like a skill from the inside.

You're young enough to have no habits to unlearn. People who've done something one way for twenty years find it hard to see that the way itself has changed. You've no such problem.

You've seen what a finished thing looks like. Most people your age have used AI to write an essay. You've watched a working product get built in an evening. That's a different mental model and it's worth more than it sounds.

How this usually starts

Not with a business. With a favour.

Someone mentions something tedious. You say "I think I could sort that". You do it in an afternoon. They're genuinely surprised. They tell someone.

That's it. That's how nearly all of it begins, and it can't be shortcut.

The one thing to get right: charge the second time. If you do favours forever you have a hobby. There's a real pattern where people give work away because it felt easy, and then can't start charging because they've set the price at zero.

I'll be direct, because it's a mistake I've watched happen: easy for you is not the same as low value to them. The camera guide took an evening. That doesn't make it worth an evening.

What tends to be worth paying for

Notice that none of those say "AI". That's the point.

What I'd actually do in your position

Nothing entrepreneurial. Not yet.

Get properly good at level 4 so making a working thing is genuinely easy for you. Weeks, not years.

Then do three favours for three different people with three different problems.

Then look at what they had in common. That's your answer, and no amount of thinking in advance produces it. It has to be found.

If you want the shortcut version: the constraint is never ideas or capability. It's knowing someone with a real problem. That's a people thing, and it's the part that no tool makes faster.

Learned the hard way, more than once.

08What Won't Change

Everything in this chapter should still be true when today's tools look quaint. If you remember nothing else, remember this page.

The nine that last

1. Whoever describes the problem best gets the best result. True of AI, true of builders, true of doctors. The tools change, this doesn't.

2. You are responsible for what you send out. "The AI wrote it" is not a defence and never will be. If your name's on it, it's yours.

3. Anything factual, check. Especially numbers. Especially when it's what you hoped to hear.

4. First drafts are cheap now. Judgement isn't. When making a thing costs almost nothing, the value moves entirely to knowing which thing to make and whether it's any good.

5. Iteration beats inspiration. The person who goes round four times beats the person with the better first idea. Reliably.

6. Context is the whole game. Almost every disappointing answer is a missing-context problem, not a capability problem.

7. Taste doesn't come free. These tools produce competent and generic unless pushed. Knowing the difference is a human skill, and it's built by paying attention to work you admire.

8. The constraint is finding a real problem. Not tools, not skills. Always someone with something that genuinely hurts.

9. Be the one who checks. As it gets easier to produce plausible things, the rare and valuable thing becomes being the person whose work is actually right. That's a reputation, and it takes years to build and an afternoon to lose.

The one that matters most

Don't outsource the thinking.

Use it to draft, to argue with, to check, to build. But the deciding is yours, and if you hand that over you'll slowly get worse at it without noticing.

The tell is when you stop being able to say why the answer is right. If you can't explain it in your own words, you don't know it. You've just got a plausible paragraph.

That's the trap, and it's a quiet one.

09The Disposable Chapter

🔴 Everything below rots. Written 2 August 2026. Correct on the day, probably stale within a year, definitely stale within three.

This is the only part of the guide that needs replacing. Everything else stands on its own. If something here contradicts what you see in front of you, believe your eyes and tell me so I can update it.

Which tool

Broadly there are a handful of serious ones and they're closer to each other than the arguments suggest. Pick one and get good at it. Switching around is how people stay mediocre at all of them.

What actually separates them for your purposes:

Ask whichever you're using: "what can you actually do that I might not realise? Can you search? Can you build things? Do you remember our previous conversations?" It'll tell you, and the answer will be more current than this page.

What it costs

The serious ones are broadly the price of a couple of coffees a month, and there's usually a free tier that's enough to learn on.

Don't pay until the free version is annoying you. When you find yourself hitting limits on something you're genuinely using, that's the signal. Not before.

The one setting worth knowing

Most of them have some version of "think harder before answering". Names vary. It's slower and better for anything with reasoning in it, and pointless for simple lookups.

Find it in whatever you're using. It's usually the difference between a mediocre answer and a good one on hard questions, and most people never touch it.

How to keep this chapter current yourself

Every few months, ask whatever you're using:

What has meaningfully changed in what you can do over the last six months? Focus on things that would change how a non-technical person uses you day to day, not on benchmarks.

That question ages better than anything I could write here.

Made for Chloe, August 2026.
Back to the top

The tools chapter goes out of date on purpose. Everything else shouldn’t. If something here contradicts what you see in front of you, believe your eyes and tell Grandad.