Descriptions:
Mikhail Parakhin, CTO of Shopify and former CEO of Microsoft’s Bing, Edge, and Windows business unit, joins the Latent Space podcast for a technically dense conversation about how Shopify is rebuilding its engineering and ML operations around AI agents. The discussion centers on Tangle and Tangent, Shopify’s internal agentic ML pipeline tools that let engineers — and increasingly product managers — compose, iterate, and deploy machine learning pipelines through natural language and YAML-based component definitions, without manually modifying underlying code.
Parakhin shares internal adoption metrics showing daily active AI tool usage as a percentage of the entire Shopify workforce, along with concrete examples of agents uncovering engineering waste: one found that a major database table was simply hashing one random ID into another, enabling its elimination. He also describes Shopify’s auto-research system, which uses agents to iterate on ML pipeline components, and explains why it has spread virally internally — with a PM now ranking as the organization’s top user. On agent architecture, Parakhin argues the key anti-pattern is running too many parallel agents without communication, and that smaller numbers of agents with structured critique loops — where one agent produces output and another critiques it, ideally with a different model — dramatically improve code quality despite higher latency.
The episode also covers Shopify’s use of Graphite stacks for managing high-velocity PR workflows in an AI-accelerated engineering environment, pro-model PR review, and the human-in-the-loop checkpoints that Shopify builds into its agentic systems. For engineering and AI leaders navigating large-scale AI adoption, this episode offers unusually candid, data-backed operational insight.
📺 Source: Latent Space · Published April 22, 2026
🏷️ Format: Interview
