Written by Piotr Kuczyński with Claude (Anthropic) as writing tool.
The Judo Portal: A Cautionary Tale
Two years ago, my son started competing in judo. I found myself sitting in tournament stands, refreshing a confusing bracket on my phone, trying to figure out: Is he still in? Who does he fight next? What’s this “repechage” thing, and does it mean I should be cheering for the kid who beat him?
The existing tools were terrible. JudoManager shows brackets, but it doesn’t translate them into what a parent actually needs to know: is my kid done for the day, or should I stay?
So I started building. A simple filter by club. A status tracker. Natural language explanations of repechage logic.
Within weeks, I had:
- Dark mode and light mode
- Separate profiles for parents, coaches, and athletes
- A historical analysis system for club performance
- Plans for video integration
- A half-built notification system
What I didn’t have: anything I could actually use at the next tournament.
I couldn’t edit myself. Every feature seemed necessary. Every addition felt like progress. But progress toward what? The original problem — “is my kid still fighting?” — was buried under an architecture designed for a platform I’d never ship.
TripForge: Where the Method Was Born
When I started TripForge — an AI travel planning tool — I recognized the pattern beginning again. Chat interface. Location validation. Weather integration. Trust hierarchies. The feature list was growing.
But this time, I tried something different.
Instead of building alone, I started staging discussions. Not with other humans — I was a solo builder — but with AI, orchestrated to represent different perspectives. A product strategist pushing on scope. A UX specialist questioning the interface. A skeptic asking why anyone would use this instead of just googling.
These weren’t casual chats. They were structured debates with clear questions, multiple viewpoints, and a facilitator synthesizing insights. The format forced me to articulate assumptions I’d been making unconsciously. It surfaced tensions I’d been avoiding.
And something unexpected happened: I started editing.
The discussions gave me permission to cut. When a panel of experts (simulated, but rigorous) agreed that a feature was premature, I could let it go. When they pushed on whether the core experience was working, I focused there instead of adding more.
TripForge shipped. Not as a sprawling platform, but as a focused tool I actually use for my family’s travel planning. The chat-first interface works. The trust hierarchy makes sense. The three-mode structure (dreaming, planning, traveling) emerged from those discussions, not from my initial feature list.
The Pattern Across Six Projects
Looking back across everything I’ve built — the shopping list app, the judo tools, the community platform concept, TripForge, the methodology itself, this very page — I can see two distinct eras:
Before structured discussions: Feature creep. Scope expansion. Unshipped projects.
After structured discussions: Clarity. Shaped bets. Working products.
The difference isn’t discipline or willpower. I have no more of those than I did before. The difference is external structure. A process that forces me to articulate, defend, and ultimately edit my ideas.
The projects that shipped aren’t the ones where I had the best ideas. They’re the ones where I subjected my ideas to structured debate.
Why This Matters Beyond Personal Projects
I spend my professional life helping organizations navigate AI product decisions and transformation challenges. I’ve facilitated workshops with 40+ executives. I’ve sat in rooms where smart people couldn’t agree on priorities, where every initiative seemed urgent, where the feature list kept growing but nothing shipped.
They have the same problem I had. They can’t edit themselves.
Not because they lack intelligence or experience. Because they’re inside the system. They can see the trees but not the forest. Every feature has a champion. Every initiative has a sponsor. The editing function — the ability to say “not this, not now” — is absent because no single perspective can see the whole.
What works? Structured discussion. Multiple perspectives. A format that surfaces tensions and forces choices.
It’s what I learned building TripForge. It’s what I practice in the Laboratory. It’s what I offer to clients.
The Thesis
Everyone’s talking about AI agents that code autonomously. Swarms that build without human intervention. Vibe coding that turns prompts into products.
I’ve used all of these. They can help. But they can’t do what structured discussion does: create the editing function that transforms a pile of features into a product that ships.
A good discussion will always beat solo execution.
Not because execution doesn’t matter — it does. But because execution without editing produces bloat. And editing requires perspective you can’t generate alone.
The judo portal taught me what doesn’t work. TripForge taught me what does. Now I’m rebuilding the judo portal with the method I discovered — and documenting the journey here.
Next: Episode 2 — The Origin. TripForge deep dive: the 7 structured debates, 14 shaped bets, and 80+ commits that became a working product.
Attribution
The insights in this essay emerged from structured discussions with AI-simulated perspectives, drawing on frameworks from David C. Baker (expertise positioning), Ryan Singer (shaping and appetite), Paweł Tkaczyk (narrative structure), and others. The method described here — orchestrating multi-perspective debates to arrive at insight no single viewpoint can reach — is what the Laboratory practices.