Handoff Stories

Daniel: I made Wild Reader for my three- and five-year-old boys to help them with early literacy. It's a phonics tool with little games for matching letters, then words, then faces with names and reading sight words.

One fun thing: my five-year-old is pretty critical of the mobile user experience and really pushed me to make that part better. There's a reward for getting five questions in a row correct—you can create an image with Gemini's Nano Banana. The kids love making things like "a monster truck with tires made out of cakes." Previously I'd let them use my phone to make images with AI—I was the gatekeeper. Now the system gates it: five correct answers earns an image. That's a handoff: setting ground lines so I don't have to say "let me have my phone back."

Another handoff: audio dictation for non-readers. My three-year-old can't read yet, but he can speak his image prompt into the tool.

Daniel: My kids like to borrow Hot Wheels cars from my mom. They can only borrow one—it's a lesson in responsibility. They can loan her a toy and then borrow one of hers. She has a lot.

I made a comment: "Maybe if we made a library and took pictures, like a lending library, then we could better track what cars you're using." The fun part was the hands-on experience with my five-year-old—taking pictures of the cars and giving them names together. Building the tool was part of the point.

The kids don't use it independently—they don't have phone or computer access. The visibility is for me and Popmom.

Daniel: A baseball positioning game built with a 10-year-old nephew. You choose a position, get a scenario (runner on second, two outs, you play first base), and when the ball is hit you have three seconds to click where you're going to go.

The fun part was the brainstorming and design thinking—helping him identify what he wanted to build and how to make it work. The goal is to help you think about what you should be doing—to ingrain the knowledge before you're on the field.

Daniel: This one started from helping a brother-in-law manage his adult soccer league schedules. Complex constraints—multiple teams, field availability, player preferences. AI seemed promising, but the results were inconsistent. That inconsistency became interesting.

So I turned it into an exploratory benchmark. Each model receives the same long, messy natural-language scheduling request. Each produces an HTML solution with a proposed schedule and explanation. Then every model reviews every other model's work and answers: "Is this solution correct? Why or why not?"

There's no ground-truth checker. No enforced rubric. The models generate the solutions and the critiques entirely on their own.

Daniel: Built with my niece, who wants to have a small ranch or farm. She needs to learn how to work on her house, truck, fence line, animals—small ag work.

Rather than explaining to a chatbot each time or working within the confines of a chat system where you repeat your searches and save documents elsewhere, we built a tool that remembers her context. She inputs a project, gets dynamic questions, sees her previous context, and the AI suggests updates she can accept or reject. Then it prepares a report she can discuss and refine.

Daniel: Built with my 13-year-old nephew who has a YouTube channel for stop motion animation. The tool uses LLMs to help plan the shots—generating scripts from ideas and shot lists with ASCII sketches.

We built a tool with AI-generated scripts and shot sketches. But we don't yet know if those outputs would actually help his workflow. He's made many videos—his process is practiced, embodied. The handoff attempt surfaced what we need to learn: his tacit process, what he actually needs.