topcut.ai

Most of my work runs through an institution. This one doesn't. topcut.ai is the control experiment: no campus, no committee, no shared infrastructure. If I can only build inside a large organization, that's worth knowing. Turns out I don't need one. The method carried over, though: it was built the way I teach building, AI agents doing the heavy lifting, me directing and reviewing every change.

The domain is deliberately fun. Competitive Pokémon TCG has real tournament data, a live metagame that shifts every weekend, and players who want an edge. That's a strategy problem wearing a card game costume. I like strategy problems.

topcut.ai layers the part players actually want on top of the raw data: analysis. Four pieces:

• Live Meta. The whole metagame at a glance: archetypes from recent tournaments, ranked by real win rates and play share. Open any archetype for decklists and matchup spreads. Not vibes; event results.
• Matchup Coach. Pick your deck and an opponent's. A panel of expert AI models reads both lists and returns a win-rate estimate, tech-card swaps, and a turn-by-turn game plan.
• Deck Builder. Start from a blank slate. Say how you like to win, pick your star Pokémon, and the lab seeds a legal list around it, then grades the deck as it grows. The grade is the feedback loop.
• Card Browser. Every in-format card, 3,000+, searchable as you type. Open any card for full text, synergies, and counters.

The tagline inside the app: "The Meta, charged."

The Matchup Coach runs a panel of models, not a single oracle. That's not a feature idea; it's a habit. My day job includes multimodal safety evaluations: put the same problem in front of five models and study where they disagree. I don't trust one model's opinion unchecked at work, so topcut.ai doesn't either.

Raw data is a commodity. pokemontcg.io has the cards. Limitless has the tournaments. The gap is between "here is the data" and "here is what to do Saturday morning." topcut.ai lives in that gap: matchup spreads ranked by real win rates, AI-generated deck synopses and game plans, deck-building guidance grounded in the current tournament meta. The mechanism is simple: real event data in, structured strategy out.

A side project becomes a product when the unglamorous parts exist. They exist:

• Topcut.ai LLC. A real entity, not a hobby URL.
• Terms, privacy, and refund policies written and published before launch, plus the required disclaimer: topcut.ai is an independent fan-made tool, not affiliated with Nintendo, The Pokémon Company, or Creatures Inc.
• A beta funnel that does discovery. Applicants tell me where they play (TCG Live, local leagues, regionals, with their kid) and what they want first: winning tough matchups, building better decks, keeping up with the meta, coaching their kid. The answers will shape the build order. Customer development, one modal at a time.
• llms.txt, robots.txt, sitemap. The site explains itself to AI search, because that's where discovery is going.

Anyone can demo a feature. Fewer people incorporate, write refund policies, and instrument their signup flow. The boring parts are the tell.

Deployed, invite-only beta; applications open. The app is live; the marketing site carries recorded demos of all four features. Launch plan: a free tier plus one-time AI-analysis credit packs: Starter, Standard, Pro. Pay for the analysis you use; no subscription.