04 / Shop · Retrospective
Owl Park
storefront.
A live mock e-commerce store with three reactive agents. Now retired — the writeup is below, the play demo is at the bottom.
Retired · 2026A reactive market in miniature.
The Owl Park storefront was a real working store backed by Supabase. Real product rows, real prices, real inventory, real cart and checkout flow.
Behind it sat three n8n agents specialized to be different parts of a live ticket market. A pricing agent watched demand and adjusted prices inside fairness bounds. A restock agent pulled units from a virtual warehouse when stock crossed a threshold. A fair-buyer agent compared prices to nearby comps and bought when the market was favorable.
Every action a visitor took — adding to cart, completing a checkout, abandoning at promo entry — fed back into the agents' state. In other words, your decisions directly influenced the agents and pricing!
The agents found their rhythm.
The pricing agent settled into a reasonable diurnal pattern within 48 hours — leaning prices up during morning peaks, easing down in the late afternoon as ride passes saturated. Without any explicit time-of-day rules — it just noticed.
The restock agent was the boring one, in a good way. Threshold checks fired on schedule, replenishments completed cleanly, never a stockout during the run. It was the bottleneck in the end, not restocking often enough to keep up with a high demand market.
The fair-buyer was aggressive during this project. I found that I could manually set prices that slowed it down, but it appeared my pricing agent was aiming for higher volume over price per ticket.
— Project notes, day 9Cart abandonment on the promo-code field might be interesting to explore in the future.
Three things I uncovered.
Reading the data afterward in Power BI was the part I'd repeat first. Watching the agents in real time was fun, but the post-run dashboard — basket size, abandonment funnel, agent-driven price elasticity — is what made this project useful.