We let AI agents make a children’s book, and kept humans in charge.

I did not start Story Genie because I wanted to make a children’s book. I started it because I had a question I could not answer from a whiteboard. So I built a whole company to answer it.

After 15 years of design leadership, including running an AI-first transformation inside a multi-billion-dollar eCommerce design org, I kept circling the same question. Can AI agents do the real production work of a creative product and still hand a person something they are proud to give? Not a demo. Not a toy. A real gift, for a real kid, paid for with real money.

There was only one honest way to find out. Build the whole thing and ship it to real people. So I did. I built it end to end. The brand, the UX, the web, and the agentic production pipeline that actually makes the books. Then I put it in front of parents and watched what happened.

The product, in one breath

Story Genie makes personalized hardcover children’s books. You tell us about a child. AI agents write the story and illustrate it. Humans supervise the work, set the bar, and approve the quality before anything reaches you. You see the entire finished book in about 60 seconds. The whole thing, cover to last page. Then, and only then, do you decide if you want it.

You can see the live version at Story Genie if you want the short tour. The lesson underneath it is worth slowing down for.

That last part, seeing the whole book first, sounds small. It is the most important decision I made. And it is the one that taught me the most about designing trust into an AI-native product.

The hard problem was never the screens

I am a designer. My instinct, walking into this, was to make the screens beautiful. Clean flows, nice type, a delightful preview. All of that matters, and I built all of it. But the screens were not the hard problem.

The hard problem was trust.

Think about what we are actually asking a parent to do. Hand over their child’s name. Hand over a little piece of who that child is, their world, the people they love. Then trust a machine to turn it into something worth keeping on a shelf for years. That is not a casual ask. A parent’s guard is up, and it should be.

No amount of rounded corners fixes that. Trust is not a visual style. It is not a color palette or a friendly mascot. It is a promise you keep, made visible in the product itself. I could make the prettiest preview in the world and it would not move a careful parent one inch if the underlying deal still asked them to gamble.

Once I saw the problem clearly, the whole design changed. I stopped asking how do I make this look trustworthy and started asking how do I make this actually be trustworthy, and then show it. Those are very different questions, and only the second one leads anywhere real.

Why we show the whole book before charging a dollar

Most of this category works the same way. You get a blurred preview. A teaser. A few pages, then a wall. Pay first, see the rest later. The whole model runs on hope. You are trusting the company before they have earned a thing from you.

I hated that the second I saw it. Not because it is unfair exactly, but because it is backwards. It asks the customer to carry all the risk at the exact moment they know the least. The company holds all the information and all the leverage, and the parent is asked to leap anyway.

So we flipped it. You see the entire finished book before you pay. No blur. No locked pages. No trust-me checkout. Every page, the words and the pictures, done and real. If it is not good, you walk away and it cost you nothing but a minute of your time.

This was not a marketing trick. It was the design solution to the trust problem. Instead of asking parents to believe us, we show them. The product makes its own case. We moved the proof to the front, before the money, instead of hiding it behind the money. The risk shifts off the customer and onto us, which is exactly where it belongs.

It is also terrifying as a founder. It means the book has to actually be good every single time. There is no clever copywriting to paper over a weak result. There is no fine print to lean on. The work has to carry it, on its own, with the customer looking right at it. That pressure is the point. It keeps us honest in a way a pricing page never could. A blurred preview lets you ship something mediocre and still get paid. Showing the whole thing does not.

Where the AI does the work, and where it does not

Here is the line I care about most, because it is the line everyone gets wrong.

The AI agents do the production. They write. They illustrate. They do the heavy creative lifting that used to take a team of people weeks. That is real, and it is the reason the whole thing can exist at the price and the speed it does. I am not shy about that part. Pretending the machine is not doing real work would be its own kind of dishonesty.

But the agents do not get the final say on quality. Humans do. Always.

This is not a hedge or a feel-good line for a landing page. It comes from a hard technical truth I learned the slow way. Models can be confidently wrong. They will hand you something that looks polished, reads smoothly, and is quietly broken. A name spelled wrong. A tone that drifts off the rails. A scene that technically matches the prompt and completely misses the heart of the story. The model has no idea it failed, because looking right and being right are not the same thing to a machine. It cannot feel the gap.

So humans stay on the bar. We define what good means for this product. We catch the misses the model cannot see in itself. We approve, or we send it back to be redone. The agents are fast and tireless and genuinely creative. They are not the judge. We are. And the moment that stops being true, the product stops being something I would put my name on.

I wrote more about this split in a separate piece on why AI replaces the process, not the people. The short version is this. The work changes. The judgment does not. Somebody still has to know what good looks like, and care enough to demand it, and that somebody is a person.

What agentic product design looks like when the stakes are real

There is a lot of writing right now about AI agents doing things. Most of it is about tasks where a mistake is cheap. Draft an email. Summarize a doc. Sort a list. If it is wrong, you shrug and fix it and move on with your day.

Story Genie is not that. The stakes are real. A real parent. A real gift. Real money out of their pocket. A keepsake that is supposed to mean something to a small person who will remember it. When you design an agentic product where the downside is a disappointed kid, every easy assumption gets stress-tested fast. Here are a few things I learned that I did not expect going in.

Speed only matters after trust is solved. A 60-second preview is a great feature. It is worthless if the person does not believe the result. I had the order backwards at first. I thought fast would feel impressive on its own. It does not. Fast only feels good once the customer already trusts that fast also means good. Before that, fast actually reads as suspicious, like a corner was cut. Solve trust first, and then speed becomes a gift instead of a red flag.

Transparency is cheaper than persuasion. I spent real energy early on trying to convince people the books were good. Better words, better claims, better proof points. The thing that actually worked was simpler and cheaper. Just show them the whole book. Showing beats telling every time, and it costs less to build. When the product is genuinely good, the most efficient marketing in the world is to get out of the way and let people see it for themselves.

The human in the loop is a feature, not a cost. It is tempting to treat human review as overhead, the slow expensive part you optimize away as the models improve. I see it the exact opposite way now. The human bar is the brand. It is the whole reason the work can be trusted. Take the humans out and you do not have a leaner product. You have a worse one that nobody careful should trust, dressed up to look the same.

Automate the labor, never the standard. This is the one I would tattoo on the wall. The agents carry the labor, and they carry a staggering amount of it. The standard belongs to people, and it should stay that way no matter how good the models get. The moment you let the model define what good is, you have handed your quality to a system that cannot tell when it has failed. The labor is the model’s job. The standard is yours, forever.

The thing I keep coming back to

When people hear AI-native company, they often picture a product where humans got removed. The robots took over and the people left the building. That is not what we built, and I would argue it is not what good AI-native means at all.

We built a product where AI does an enormous amount of real work, and humans stay exactly where humans belong. On the standard. On the bar. On the one question that actually matters, which is whether this is good enough to give to a child. The machine answers the how. The people answer the whether. Both jobs are essential, and they are not the same job.

Why radical transparency is a moat, not a marketing line

Most companies in our space hide the work. They show a blurred page, a watermark, a teaser, and then they ask you to pay before you can see what you actually get. That choice feels safe to them. It protects the output from scrutiny until the money is already gone. We went the other way. You see the whole book first. Every page, every illustration, the real words your child will read at bedtime. Then you decide. I want to explain why that is not just the kind thing to do. It is the strongest competitive position we have.

A moat is something a competitor cannot copy without paying a real price. Showing the whole book before charging is exactly that. If a company built on blurred previews tried to match us, they would have to reveal their full output too. And the moment they do, their customers can compare. The reason they blur is almost never mystery for its own sake. It is because the finished thing does not hold up under a clear look. The blur is doing a job. Remove it and the weakness shows. So they are stuck. They can keep hiding and look less trustworthy than us, or they can reveal and expose work that was never built to survive a close read.

That is the trap, and it is a good one. Our transparency is cheap for us and expensive for them. It is cheap for us because we built the book to be seen. Every part of the pipeline was designed knowing the customer would judge the final result with nothing hidden. It would be expensive for a blurred-preview company to match because they would have to rebuild their quality bar from the ground up first, then change their whole funnel, then retrain their customers to expect honesty. You cannot bolt transparency onto a product that was made to be hidden. You have to mean it from the start.

Here is the part I care about most. Showing the whole book first is a forcing function on the entire company. When the customer sees everything before paying, there is no place for a weak page to hide. A clumsy sentence, an illustration that misses, a name spelled wrong, a story that goes flat in the middle. All of it is right there in plain view at the exact moment someone decides whether to trust us with a gift for their kid. That pressure flows backward into every team and every agent in the pipeline. The standard is not a rubric on a wall. It is the customer’s own eyes, every single time.

This changes how we work in a real way. We cannot ship a book that is good enough on average and count on a blurred preview to carry the weak ones across the line. Every book has to stand on its own, fully visible, before a dollar moves. That is a harder bar, and it is the right one. It pushes the AI agents to write tighter and illustrate cleaner, and it pushes the humans who supervise them to approve nothing they would not hand to their own child. The transparency and the quality are not two separate goals. One enforces the other.

So the moat and the discipline are the same thing seen from two sides. From the outside, radical transparency is a position competitors cannot copy without hurting themselves. From the inside, it is a promise that keeps us honest and keeps the work genuinely good. We do not show the whole book because it tests well. We show it because it makes us better and it makes us hard to follow. The day a customer can see everything and still chooses to pay is the day you know the product earned it.

The goal was never a book made by AI. It was a book a parent loves, that AI happened to help make.

That sentence is the whole philosophy in one line. The AI is the how. The parent’s love is the why. The day you confuse the two, you start building something that demos beautifully and disappoints in the hands of a real person, which is the worst possible place to find out you got it wrong.

If you are building an AI-native product where the stakes are real, that is the line I would hold onto. Let the agents do the work, all of it they can handle. Keep the people firmly on the standard. Show your customer the truth before you ask for their money, and then do the harder thing, which is making sure the truth is actually worth showing.

And if you are wrestling with where exactly to draw that line in your own product, that is the kind of problem I love to dig into. You can read more about how I think through it in my AI design consulting work, and you can see the full story of how this one came together in the Story Genie case study, messy parts left in.

Build it so you would be proud to hand it to someone you love. Then let the machine help you do exactly that, at a speed and a scale one person never could reach alone. That is the whole bet. So far, it is holding.