CI/CD Breaks at AI Speed: Tangle, Graphite Stacks, Pro-Model PR Review — Mikhail Parakhin, Shopify

From running one of the most aggressive internal AI rollouts in tech to building systems that simulate customers, optimize ML pipelines, and rethink search latency from first principles, Mikhail Parakhin is pushing Shopify far beyond “add AI to ecommerce.” In this episode, Shopify’s CTO joins swyx to unpack what actually changed inside the company after the December 2025 model inflection, why nearly everyone at Shopify now uses AI tools daily, and what happens when code generation outpaces the rest of the software delivery pipeline. We go deep on Shopify’s internal AI stack: Tangle for reproducible ML and data workflows, Tangent for auto-research loops that keep iterating toward better results, SimGym for customer simulation and theme optimization, and Liquid AI for ultra-low-latency and long-context workloads. Mikhail also reflects on PR review in the age of coding agents, the future of CI/CD, Sidekick Pulse, UCP, hiring at Shopify, and the untold story behind Sydney at Microsoft. We discuss: • Why Shopify’s AI usage effectively reached company-wide adoption • The December 2025 inflection when model quality suddenly changed behavior • Why token budgets matter, but only if paired with strong critique and review loops • The real bottleneck in AI coding: PR review, CI/CD, and deployment stability • Why more AI-written code can still mean more bugs in production • Tangle: Shopify’s reproducible workflow system for ML and data experimentation • Tangent: auto-research loops for optimizing pipelines, prompts, throughput, and infra • Why auto-research is democratizing ML experimentation beyond researchers • SimGym: simulating customers using Shopify’s historical data and browser-based agents • Why customer simulation is hard to copy without years of real merchant behavior • HSTU, counterfactuals, and modeling merchants and buyers over time • UCP and Shopify’s product catalog as runtime infrastructure for agents • Why Shopify is using Liquid AI for low-latency search and