LunchBox Buddy

LunchBox Buddy is a small AI tool aimed at one of parenting's most relentless chores: packing school lunches. The flow is simple — open the app, snap a photo of what's in your fridge or pantry, and LunchBox Buddy returns suggestions for what to actually pack that day. The AI takes inventory of the food it can see, considers some basic kid-friendly lunch heuristics (a protein, a fruit, a snack, something the kid will actually eat), and turns that into concrete suggestions instead of leaving the parent staring at random ingredients at 7am.

It's a tiny, very specific use of multimodal AI, and that's the point. The friction in packing lunch isn't ingredient access — it's the daily decision-making load. LunchBox Buddy collapses that into a single photo.

How it was built

LunchBox Buddy was vibe-coded by Kevin Roose, a New York Times technology journalist, who wrote about the experience as part of his ongoing reporting on AI tools. The app was built using Bolt as the primary scaffolding tool, with a stack reportedly built on Nuxt and daisyUI / Tailwind for the front end. It runs as a mobile-friendly web app rather than a native binary.

Roose's piece is one of the more widely-read examples of a non-engineer journalist shipping a real, working product as a way of stress-testing what AI tools can actually do. The app is also catalogued in the awesome-vibecoded-apps list on GitHub.

Why it matters

LunchBox Buddy is a clean example of the genre of "personal tool a parent builds for themselves and then puts on the open internet." Five years ago, doing this would have required either a willing engineer in the family or a multi-week side project; with Bolt and Claude under the hood, it became a weekend's work. The fact that the builder is a tech journalist makes the artifact double as reporting — proof of concept and case study in the same product.