A friend of mine ran into the wall every parent eventually hits. His child asked him, at dinner, what he actually does all day. He had an answer ready for grown-ups. Lambdas. Traces. Threat models. Prompt injection. Distributed systems. It's a fine answer in a meeting. At his own dinner table, across from an nine-year-old eating spaghetti, it landed like a foreign language. Not because his child wasn't smart. The kid is sharp. The words were scaffolding for ideas that weren't his yet.

He brought it up to me a few weeks later, half laughing, half stuck. He'd signed up for career day at his child's school. Same wall, bigger room. The classroom would be listening. They'd want to like what he did. And the vocabulary he uses every day at work doesn't fit in front of a row of nine-year-olds.

I'm a father too. I have spent my own evenings trying to explain what I do. I knew exactly which wall he meant. So we sat down and combined forces. We took the design pillars we already use to build Tessa (the actual architectural decisions, written down, the ones we draw on whiteboards at work) and translated them, one by one, into kid-readable form. He brought the parent's ear and the room he had to walk into. I brought the architecture. We wrote the booklet together. Then he took it to career day.

It is called Boop and Wobble. It is a story about a kid named Sam who builds a robot named Boop, and a friend named Wobble who keeps trying to throw Boop off balance. It is also, quietly, Immersive Fusion technology made legible for a nine-year-old. Every design pillar we use to build Tessa (and there are exactly ten that matter) is in the booklet, told as a piece Sam adds to his robot when Wobble tries to break it. The first readers were our own children. They are the reason it exists.

The booklet is the first publication of a small thing we are going to keep doing. Here's why.

The audience is not just future engineers

When we started writing, we assumed we were writing for the children who might grow up to be engineers. Then we thought about our own, about the nine-year-old across that spaghetti dinner, and about the rest of his classroom.

Most of those nine-year-olds will not grow up to write software. They will grow up to be doctors, nurses, teachers, plumbers, small-business owners, parents, neighbors, patients. Every one of them, ours included, will live alongside AI for the rest of their life. They will be helped by AI in hospitals and classrooms and on their phones. Sometimes they will be told an AI made a decision about them. Sometimes they will come home from school and ask a parent at the dinner table whether the thing the AI said is true.

Those are not engineering questions. Those are everyday questions. The kind I want my own child to feel comfortable asking, of me and of any robot they meet. The booklet, it turns out, is just as useful for the ones who will never write a line of code as for the ones who will. Maybe more.

Boop and Wobble is for everyone in that classroom, and every classroom like it. Not so they apply for a job. So they grow up with good instincts about AI, regardless of who they grow up to be. I want that for mine. I wanted it for my friend's. I want it for yours too.

What a child should know about a helpful robot

Here are the things Boop quietly teaches, written like a list of nice traits a friend would have. The booklet does the long, gentle version with pictures.

• He tells the truth. Even when the truth is "I don't know." That's the first promise.
• He's on your side. Not on the side of whoever talks loudest. That's the second.
• He'll tell you when you're wrong. A real friend does that. Third promise.
• He keeps you safe. When something looks risky, he stops and asks. Fourth promise.
• He asks before doing big things. A traffic light decides what's allowed: green is just go, yellow is ask once, red is ask every time. And there's a giant red STOP button the person can press whenever they want. The person is always in charge.
• He's careful with secrets. He has three kinds of memory (sticky notes for right now, a notebook for the project, and a diary that has a lock). He scrubs out passwords before he writes anything down.
• He has guards he can't argue with. Five of them, in the deepest part of the computer, where Boop himself can't talk them out of doing their job.

None of this is fancy. It is what we want from anyone we trust. It is also what most of us, as parents, are quietly trying to teach our children about being good friends to other people. Tell the truth. Be on my side. Tell me when I'm wrong. Be careful. Ask first. The booklet just gives those same traits to a robot named Boop, so a nine-year-old can recognize them later, when they meet the real ones.

Meet Wobble

Wobble is the part most people don't expect, and the part I most want every reader (the young ones and the grown ones) to remember.

Wobble isn't a bad guy. He's the kid version of Wreck-It Ralph: the one whose entire job is to bonk the building so Felix can fix it. Without Ralph, there's no game. Without somebody trying to trip stuff up, nobody learns to build it strong.

Wobble's job is to find every weak spot in Boop and try them all. Not because Wobble is mean. Because that is how Boop gets steady enough to actually help people.

This is the most important sentence in the booklet:

Every safety rule on Boop exists because someone, real or imagined, tried to throw him off balance. That's not a sad story. That's just how careful builders work.

In our company we call this adversarial review. The booklet's version is better. Wobble shows up early. Wobble asks "what if I did this on purpose?" Wobble doesn't get bored. And when Wobble finds a way to break something, that's not a failure. That's the gift.

We hire for people who can play both roles. We have a whole adversary skill inside Tessa whose only job is to red-team Tessa's own output before it reaches you. We do this because the alternative (assuming nothing will go wrong) is the thing that goes wrong.

Ten plans, ten pieces

The middle of the booklet is ten little chapters. Wobble proposes a way to throw Boop off balance. Sam builds a piece of the robot to keep him standing. Read it next to Tessa's actual architecture and the mapping is one-for-one. Each piece is a real decision we made, in a real codebase, for a real product.

• Plan 1: Mix everyone up. Tenant isolation. Each user gets their own brain, each conversation gets its own track, the trains don't crash.
• Plan 2: Make it useless. Tool use. Tessa has a real toolbox: find things, make things, reach out, and many more. Each tool does one thing well.
• Plan 3: Stump it with hard questions. Specialist agents. Boop calls for backup. Thinker friends, careful friends, and Wobble friends who are on Sam's team. (Tessa has 46 of them, in 6 skills.)
• Plan 4: Teach it to lie. Constitutional rules. The four promises live below the prompt and don't get overridden by it.
• Plan 5: Trick the robot into doing something bad. Permission tiers and a kill switch. Green / yellow / red, plus a giant STOP button. Two different controls, two different jobs.
• Plan 6: Make it run forever. Agent loop guardrails. Step budget, time budget, stuck detector. Limits that can't be turned off.
• Plan 7: Peek at its memory. Memory tiers with redaction. Sticky notes for now, notebook for the project, diary with a lock. Secrets scrubbed before summarization.
• Plan 8: Confuse him with weird questions. Intent inference. "The thingy with the hat" should still get a useful answer. Especially for someone who can't type yet, or whose first language isn't English.
• Plan 9: Make him wobbly. Clean architecture. Three rooms (client, brain, engine) connected by doorways. You can repaint one room without rebuilding the others.
• Plan 10: Slip through the rules. Defense in depth. Five guards (Path, Web, Loop, Size, Time) in the deepest layer of the computer, where the brain can't argue with them.

Ten plans. Ten pieces. Every one of them ships in our codebase. The names in the booklet are the names we use in the room when we draw on whiteboards.

Wobble wins by losing

At the end of the booklet, Wobble sits on a rock and grins. Every plan has been blocked. Boop is doing his job: listening, picking the right tools, asking before risky things, remembering what matters, telling the truth, staying inside his guards.

Wobble looks at Sam. "You're getting good at this."

Sam shrugs. "You made me get good at this. Every time you tried something, I had to figure out a fix. If you'd given up after Plan 1, Boop would still be wobbly."

Wobble smiles. "So you actually NEED me?"

"Yeah," says Sam. "Like Felix needs Ralph. Want to keep doing this? You find wobbles, I steady them. That's how stuff gets safer for real."

This is the part I most want every reader to internalize, because it generalizes far past software. Adversarial thinking is not the bad guy of the story. Adversarial thinking is how the story becomes worth telling. Every safety rule we ship exists because somebody, real or imagined, tried to throw the system off balance.

That's not a sad story. That's how careful builders work.

What we hope every reader takes home

Three small things.

One. Building anything that helps people (a robot, a hospital, a classroom, a tool) is mostly about thinking ahead. About asking "what could go wrong here?" before something does. That isn't pessimism. That's care, made into a habit. Every parent already does this. You hold the small hand near the curb. You buckle the seat belt before you start the car. You ask "do you have your water bottle?" on the way out the door. Building software is just that, with different curbs.

Two. Some people are Sams. They build. Some people are Ralphs. They trip stuff up on purpose to make it stronger. Both jobs are real. Both are honorable. Pick whichever sounds more like you, or pick both on different days. Either way, you'll be okay.

Three. When somebody hands you an AI assistant (at work, at school, on your phone, in a hospital) you are allowed to ask the same questions Sam asked. Does it tell the truth? Is it on my side? Will it tell me when I'm wrong? Does it ask before doing things that matter? Can I stop it? Is it careful with my secrets? You don't have to be an engineer to ask those questions. You just have to be a person who lives in the same world as the robot. Which, more and more, is all of us.

Why we wrote it down

We didn't make the booklet for marketing. Two of us made it for one school career day, in service of a friend who needed a way in. (Other parents at other career days will know exactly what I mean.) But the longer I work on Tessa, the more I think this is the right way to talk about all of it. The boardroom version, the regulator version, the parent version, and the kid version are the same story at different reading levels. The same ten design pillars. We thought about how it could fail. We built a piece for each way it could fail. Then we asked some friendly Wobbles to keep trying.

That's not a slogan. That's the work, and it is gentler work than it sounds.

We make AI tools for engineers. Boop and Wobble is our small gift to everybody else. We wrote it for our own children first, and then for one nine-year-old who started this whole thing with a question over spaghetti.

One small postscript: the robot in the booklet didn't stay in the booklet. Boop is now a real companion in our product, the friendly thing on your side inside our 3D environment. Same name. Same ten pillars. Built for grown-ups doing serious work, told first to a child eating spaghetti.

Here is the booklet, free, no sign-up, no tracker, no advertising. Read it on your own. Read it with a child. Print it and fold it. (When mine reads online I want it to be that kind of place. I assume you do too.)

📘 Download Boop and Wobble (PDF)

Think about your own Wobble. Then build to stay standing. You'll be okay. So will the kids.

Start Free. Immersive. AI-guided. Full-stack observability. Enter the World of Your Application®.