NextShoe.Run

Context

NextShoe started as a year of conversations with ChatGPT about running — which shoes to rotate, when to retire them, how different shoes performed across different run types. The insight was simple: that knowledge shouldn’t live in a chat window. It should be a product.

The Problem

Runners who care about their gear don’t have a good tool for tracking shoe performance over time. Strava tracks mileage. That’s it. No system learns which shoe works for which run, accounts for how a shoe feels as it ages, or recommends rotation based on actual performance history.

The Approach

I built it. Full stack — Next.js, Tailwind, Supabase backend, auth flow, migration pipeline. Four days, solo, using AI orchestration throughout.

The interesting decision was what not to build. The obvious move was AI recommendations — feed run history into a model, get shoe suggestions back. I chose a weighted heuristic model instead. The dataset is a single user’s run history. AI would be overkill, expensive, and unpredictable at this stage. The heuristic model is faster, cheaper, and more honest about what it actually knows.

AI enters later, at the right seam: natural language run plan import, where it genuinely reduces friction. Not as a shortcut to skip the hard thinking about the data model.

The Run Planner lets users override recommendations with manual input, keeping the human in the loop.

What Happened

Working app. Live at NextShoe.run. Algorithmic learning that improves recommendations as you log more runs. Built in four days because the problem was clear, the decisions were deliberate, and the tools were used appropriately.

NextShoe is planned for integration into Supwell as the shoe tracking layer.

What This Shows

Product judgment about when AI is the right tool and when it isn’t. AI orchestration as a builder skill — not vibe coding, but directed development with architectural intent. Shipping under self-imposed constraints with no team, no budget, and no deadline except the one you set yourself.