I designed an AI camp activity for 55 kids. And I started in the wrong place.

Let's get something straight before I tell you all about this activity and my process. I really don't like space.

Most people (my kids included) think space is so cool. They love to talk about the planets, stargaze, visit museums related to space. I find space so expansive and a bit scary. There I said it. I just don't like it.

So when a local camp director reached out and asked me to design a space-themed activity that integrated AI for Space Camp this summer, I felt two things simultaneously: genuine excitement (a real camp, real kids, real stakes) and a little out of my comfort zone (because space).

I set that aside and got to work. And within about an hour, I had already made the mistake that I tell every parent (and even my coworkers professionally) not to make.

I started with AI. That never works.

Here's what I did: I opened both Claude and ChatGPT, gave them a full brief — the age range, the group size (about 55 kids), the space theme, one laptop to share — and asked them not just to generate ideas but to actually talk through them with me. I wanted a thinking partner, not a list.

We explored alien habitats, planet design, constellation discovery, something about rovers. The AI tools were doing exactly what I asked. The ideas were technically coherent. They were even kind of interesting.

And not a single one of them resonated with me.

Everything felt flat and forced. I couldn't figure out why I kept rejecting ideas that should have worked on paper. And then I realized: the problem wasn't the ideas. The problem was where I'd started. I had opened with "what AI integration makes sense for a space theme?" and worked backwards to find an activity that fit. I was reverse-engineering a learning experience from the tech, instead of designing one from what I actually wanted kids to walk away with.

That's the opposite of everything I believe.

When you use AI incorrectly, stop and start again

I told both tools to forget everything and start over. I actually typed that into the chat.

And then I asked myself what I should have asked first: What do I want these kids to learn? What skills does this activity build? And how can AI amplify those?

Now, I was on to something.

I knew I wanted kids to practice social skills upfront. They had to work through something, together, and come to a consensus. This forces them to use collaboration, negotiation, and persuasion skills. I wanted them to be creative directors, not consumers of tech. And I wanted AI to show up in the middle of the experience, not at the beginning or the end. Most importantly, I wanted AI to be the smallest part of the entire activity.

From there, the concept built itself.

And now we have the "Concept to Cosmos" group activity

Each group of campers gets a Mission Planning Guide. Before I do anything with a laptop or AI, every member of the team contributes ideas about an alien planet and the creature that lives on it. What does the ground look like? What's the weather like? What's the most dangerous place on the planet? What does the alien eat? How does it move?

The guide is structured so that every child has to contribute something, and the team has to agree before they can move forward. The majority of the thinking happens before AI enters the picture.

The planning worksheet is organized around two parts: the planet and the alien. Teams work through questions about terrain, climate, defining landmarks, dangers. For the alien design, rather than a blank "draw your alien" box (which creates an equity problem the second one kid can draw and five can't), the guide includes a voting grid where teams circle options across categories — size, skin texture, number of limbs, how it moves, what color it is — followed by open-ended fields where they can add the specific weird details the grid can't capture. (This, it turns out, is exactly how you write a good AI prompt: start with structure, then layer in the specifics.)

Once the team agrees on their design, Wednesday is when Mission Control takes over. That's the AI.

This is where I get to join! I will input the team's description, and the first artist's rendering of their alien planet appears on screen. As a group, teams discuss what AI captured correctly, what it missed, and what needs to change. They make one round of revisions and generate a final image.

Thursday and Friday are the build. This is where the activity becomes individualized and specific to each person's final interpretation. Each camper gets their team's printed image and uses it as inspiration — not a template to copy exactly — to build their own version of the alien and its world from craft materials. The physical build is individual, so every kid leaves with something they made with their own hands. Something that belongs to them.

Why following a certain order is critical in AI use

The structure of this activity is intentional, and it maps directly to the Idea → AI → Play™ framework that sits at the center of everything I design at AI for Curious Kids.

• Ideas first! Kids spend Monday and Tuesday in pure ideation mode. No screens. No AI. Just discussion, disagreement, and eventually, something they all agree on. The Mission Planning Guide exists to give that process structure without constraining it.

• AI in the middle. Wednesday's generation is a shared event. Because there's one laptop and 55 kids, the AI doesn't become an individual tool, it becomes a group experience. Everyone watches every team's prompt go in. Everyone sees the image come out. Everyone notices when the AI gets something wrong. Kids who watch six different groups prompt the same AI with six different descriptions come away understanding something important: what you put in shapes what comes out. Being specific matters. Being vague gets you something generic.

• Move from AI into play. Thursday and Friday are pure creation. Physical, hands-on, your-interpretation-of-the-thing making. The AI image is for inspiration only. It should not be considered the final product.

If you want to see this framework in action with a different activity, the Character to Costume activity follows the same arc: kids invent a superhero, AI visualizes it, they become the character through physical making. As soon as I run the space activity, I will feature the walkthrough on this site.

Most AI for Curious Kids activities are meant to happen in a day. This one is different.

When I sent the concept to the camp director, Ms. Becky, she loved it and immediately proposed a full-week schedule: brainstorming and planning on Monday and Tuesday, AI generation and critique on Wednesday, building on Thursday and Friday. I had designed the activity thinking it would happen in a single session. Spreading it across a week means the kids have real time to sit with their planet and alien before they ever see what AI does with it. It means the build isn't rushed.

My lesson: Continue to learn from mistakes

Here's what I want parents to take away from this, because it applies at home just as much as it applies to a camp activity with 55 kids:

• The mistake I made was a common one. I started with the tool and worked backwards to the activity. When AI doesn't land right with your kid — when the experience feels flat or forced or like everyone's just watching a screen — this is usually why. You optimized for AI usage instead of designing for the thing you actually wanted to happen.

• The fix is a single question. What do I want my kid to practice? What skill, what idea, what experience? And then: where does AI enhance that skill or idea?

For this camp, the answer was consensus-building. Social collaboration. The ability to hold an idea loosely enough that it can become something better when someone else adds to it. AI didn't teach those things. The Mission Planning Guide did. AI just gave them something concrete to argue about, and something visual to hold when they were done.

That's always been the goal. AI is the smallest part of the activity. The learning happens around it.

If you want to try the Idea → AI → Play™ framework at home, the activities library has activities you can run with kids of almost any age. Start with one. See what your kid does with it. Then let them tell you what to try next.

That's how the best ideas usually show up — not from AI, but not despite AI either. Somewhere in the conversation between.