Brackets, Libri GmbH Brackets, Bots & Bugs

What happens when a small football prediction app becomes a real-world experiment in agentic coding?Brackets, Bots & Bugs is a builder's diary from the development of a World Cup 2026 prediction app where humans and AI models competed side by side.This is not a polished case study about how everything went right. It is the raw, honest story of what happened when a side project met real users, live match deadlines, WhatsApp bug reports, token costs, scoring disputes, support tickets, and coding agents moving very fast.This revised and expanded edition (Version 2) captures the real-time builder journey from kickoff to the final whistle, then opens the black box with a complete post-tournament model autopsy.Inside, you will find practical lessons from building with agentic coding tools, especially Anti Gravity, including: - Why an empty folder and a vague prompt are not enough: Setting context boundaries for agents.- How real users turn a side project into production: Defensive support and timezone traps.- Why scoring rules and product invariants must be written down early: centralizing logic to prevent screen-by-screen drift.- Why token costs, quotas, and coding-agent usage need their own operating model: Separating developer usage from application costs.- How project wikis, system walkthroughs, and approval gates help keep humans and agents aligned.Plus, the complete Part II Post-Tournament Autopsy, revealing: - The Final Standings: How Claude, Gemini, GPT, and Llama compared to the best human predictors.- Confidence vs. Reality: The calibration curves of AI confidence (and why some models are overconfident while others hedge).- The Draw Blindness Finding: Why models predicted draws at a rate of only 3.4% compared to the actual 23% tournament rate.- Roster Nostalgia and Hallucinations: How models invented players or relied on retired legends when context pipelines weakened.- Cost-Effectiveness Metrics: Tracking cost-per-point to find the true signal-to-dollar ratio.- The Spain vs. Argentina Match 104 Autopsy: A deep dive into how models adjusted their rationales leading up to the final.The app was small. The lessons were not.For builders, developers, AI practitioners, product-minded engineers, and anyone curious about what agentic coding and LLM forecasting feel like when they leave the sandbox and meet the real world.