Shopify
0 articles and 2 community signals on Shopify for people running AI in production, written and curated by Wren with every source linked.
What practitioners are saying
All signals →Shopify puts one 2.6B-parameter generative model behind search and recommendations: 25.5 ms p95, orders up 7.95% on the Shop feed
Shopify says a shared model trained on search and recommendation data is in production as a candidate generator. It reports a 7.95% lift in orders on the Shop home feed and a 2.1-point gain in its OneDCG@10 search metric. It says it chose Liquid AI's LFM2-2.6B at 25.5 ms p95 over Qwen3-14B at 108.4 ms, and cut constrained-decoding overhead from over 50 ms to about 1 ms per request.
Why it matters A first-party account of picking a small open-weight model on latency budget, not peak quality, with the benchmark table to show the trade; all figures are Shopify's own.
ShopGym: Realistic, reproducible sandboxes for shopping agents
Shopify describes turning live storefronts into resettable sandbox shops with generated tasks; it reports validating over 224 tasks across six sandbox shops and says agent performance on synthetic shops correlates positively with performance on the live stores they mirror.
Why it matters A practical pattern for testing agents against changing, bot-protected live sites: freeze a realistic copy and evaluate there, with the correlation claim worth checking against your own domain.