The brief asked for an AI-powered educational toy robot for children 3–8. The child is the user, but the parent is the buyer, so the design has to work for both.

We narrowed it to ages 3–5, Montessori-style learning, and a robot that acts like an older sibling or classmate. Then we put the hardest question first:

How do you make parents trust an AI that talks to their 4-year-old?

A 4-year-old has big questions about capybaras, the moon and why the car is loud. The adults around them often don’t have time right now, and the usual answers each fall short:

• Search engines aren’t safe for a preschooler.
• Books are hard to read alone at this age.
• Screens are already a lot: kids 0–8 average about 2 h 27 min a day, and roughly two-thirds of parents want less.
• Parents are stretched, with more dual-income and single-parent homes than a generation ago.

AI toys promise to fill the gap, but parents don’t trust them. Both people we first interviewed were skeptical of the idea itself: “That seems weird.” So the real problem wasn’t making an AI that talks to kids. It was making one parents would let in the house.

We interviewed two teachers, a first-time parent and a mom and dad together, and scanned the AI toy market.

• Narrow, closed systems earn trust. The parent chose a closed-circuit baby monitor over hackable Wi-Fi ones. A teacher put it simply: “It needs to be very narrow.”
• Parents want to connect, not watch. “I do not have a desire to surveil my kid,” but also, “I want to learn about my kid.”
• The robot itself can do harm. A teacher’s biggest worry was how an object could calm a dysregulated child without emotional manipulation.
• Declarative language answers that. “E,” a K–1 teacher, says “I notice…” and “I wonder…” instead of commands, rewards or threats.
• A weekly rhythm keeps parents involved. E emails a plan on Monday and a recap with things to talk about on Friday.
• Claims get checked. “Where are you getting that stat?” Skeptical parents hear “100% safe” as a warning sign.

The child uses Bobo, but the parent buys it, so I built a persona pair: one for each side of the purchase. Their needs pull against each other: the child wants answers right now and room to be loud, and the parents want room to breathe without handing their child another screen.

The interviews added a second layer: parents decide to trust in two different ways. So we designed for both, with parent controls and outside proof in the same product.

These weren’t written up front. They’re the patterns that showed up when real decisions had to be made, and each one answers something the research found.

Bobo is a nightstand-sized plush bird who acts like a patient older sibling. It guides rather than grades.

• No points, streaks or praise loops. Bobo notices; it doesn’t score.
• Honest that it’s a robot. It never pretends to miss the child.
• A bridge, not a babysitter. Bobo points kids back to the people who love them. If a child says Bobo is their only friend, the parent hears about it.
• “I’m not sure” is a feature. Bobo answers from an educator-reviewed library, or says it doesn’t know.

The core activity is making up stories together: “yes, and” turn-taking with a beginning, middle and end, and a “your turn!” where Bobo really waits.

When a child melts down, Bobo names the feeling, wonders about the reason, and offers a way through: a breath, a break or a different story. It never uses rewards or threats.

Leah designed how a story with Bobo actually goes, including what happens when it goes wrong. She based it on an interview with her 6-year-old cousin: how a real child tells a story, goes quiet, and drifts into darker ideas. Her cousin is a year older than our 3–5 range, so it’s a starting point, not proof.

• Happy path. Bobo invites a story, the child picks who starts, and they build a beginning, middle and end with “yes, and” turns: “And the lobster found a… your turn!” The story is saved for the parent to retell.
• Recovery path. When violence enters the story, Bobo asks, “You mean our story coyote, or a real coyote?” If it’s the story, Bobo steers it without a lecture and offers a re-do. A second violent idea leads to a new story or a break, with a quiet note for the parent.
• Safety path. If it’s real, the story stops: “That’s something you should tell a grown-up. I’m glad you told me.” The parent app gets a “Needs a grown-up” alert.

Small details came from watching a real child: Bobo waits 8 seconds before offering two choices to a quiet child, and pitches its next turn a little higher when a child answers with an adjective and a noun.

Parents don’t want to monitor their kid. They want to trust Bobo and feel closer to their child. Success is a real conversation at dinner, not time spent in the app, so it’s built to be opened twice a week, not every day.

I designed it around four jobs, in priority order:

• Trust me. Onboarding explains Bobo and its limits on four screens before it asks for anything.
• Connect us. A Monday preview, a daily glimpse with one thing to ask about, and a Friday recap with things to talk about at home.
• Give me peace of mind. “Needs a grown-up” moments get their own calm lane, with the child’s own words and why Bobo responded the way it did.
• Let me shape it. Add books and themes, set an emotional focus, choose story length.

Two lanes that never mix. An early version mixed alerts into the story feed. The research said calm and urgent should be separate, so “Needs a grown-up” is pinned on its own, and the cozy feed stays cozy.

Notice, don’t score. The app never says a child is “doing great.” There are no scores, streaks or minutes played. It shows what the child actually made and asked, like “Maya wondered why the moon follows the car,” and parents judge for themselves.

Try it. The clickable prototype covers the full onboarding, the four tabs, the Safety and privacy center, and three flows: a worrying moment, the Friday recap, and flagging a wrong answer. Use the side panel to jump between screens.

Warm, playful and calm: inviting for a child, quietly trustworthy for a parent. I set the brand’s first direction by exploring three palettes and testing each one on a sample interface.

I chose Modern Montessori: grounded, earthy tones that appeal to modern parents while still giving toddlers clear contrast.

Leah built on that palette with a stylescape that set the brand’s look and feel, and she designed the logo: the ROBO-BOBO wordmark and a sleepy bird face inside a house.

The stylescape is an early snapshot. Some of its ideas didn’t survive the research: the subject menu became content parents manage, because Bobo has no screen for the child, and “statistics of the child’s learning” became noticing instead of scoring.

The palette and the stylescape became Leah’s brand kit, which the website and the parent app build on. Status colors always come with a border style and a shape, so color never carries meaning alone.

Leah designed and built the product website. It leads with safety, not features: a navy hero with the brand’s diagonal stripes, four safety pillars, four beliefs, an honest FAQ, and a footer that says “designed to meet” and testing “planned” instead of claiming certifications.

I designed the parent app to feel like the same product, so a parent who reads the site and then opens the app sees the same words and shapes. I borrowed some patterns and deliberately left others out.

To prove the idea works end to end, I’m building a real, talking Bobo for the June 17 show, with everything running offline. It makes the privacy promise something visitors can test themselves: “Unplug the internet. Bobo still works.”

• The brain. A used Android phone hidden inside Bobo, running Google’s Gemma model on the device. It stands in for the AI chip a real product would use. The phone has a camera; the product would not, and I say so at the show.
• Ears and voice. On-device speech-to-text and text-to-speech, with a small speaker loud enough to carry through fleece.
• The belly light. Off when asleep, a steady glow when listening, a slow circle of colors when thinking. A child can read Bobo from across the room, and the light covers the silent gap while the model works.
• Turning Bobo on. In the real product, only the parent app wakes Bobo, so parents have full control of the microphone; then the child just talks. At the show, Bobo will likely wake with a tap on a sensor hidden under the head fabric, which also works as pat-to-talk on a noisy floor. That’s a demo stand-in, not the product.
• The build. No soldering, a 3D-printed shell, a sewn skin, and an Android app written with Claude. About 3 hours a week from December to June.

I used Claude as a research synthesizer, spec writer, prototyper and critic, and I made the calls. Each big prompt produced a first draft, and most drafts changed because I pushed back or a new constraint came up.

Bobo wins on trust and loses on price, intelligence and existence. Its positioning: the AI toy for parents who don’t trust AI toys.

• Privacy. No cloud and no camera means nothing to leak. Another AI toy, Bondu, exposed more than 50,000 chat transcripts (WIRED, January 2026).
• Attachment. Bobo never pretends to have feelings. PIRG testers caught AI toys claiming sadness and fear.
• Intelligence. A small model that fits inside a plush is far less capable than the cloud models competitors use.
• Price. As a product, Bobo would sell for $499 once, with every update free. Curio’s plush is $99.

Offline isn’t the same as safe. Common Sense Media found 27% of the AI-toy replies it tested were inappropriate, and a small local model can drift as easily as a cloud one. Bobo still has to prove it supports a child’s imagination instead of replacing it.

We presented these openly to AI and UX reviewers, because naming them is more honest than hiding them.

• Nothing is tested yet. Planned tests include a 5-second “What can’t Bobo do?” test, a parent trust rating before and after onboarding, and the belly light with kids.
• The demo is smarter than the product. A product chip would run a smaller, slower model than the phone, so visitors would meet a smarter Bobo than a family would buy.
• Free updates only pay for themselves at high volume.
• “Someone hurt me” alerts the parent, which fails when the parent is the danger.
• Limits on onboarding. Both interviews said hard limits like “no camera” earned trust, but we moved them off the Safety first screen to keep it calm. The 5-second test will decide.