Episode Summary
Executive Summary: Shopify’s engineering leader Farhan Tavar describes a deeply AI-first operating model built around experimentation, internal platforms, and role modeling. Shopify adopted Copilot early, uses Cursor and Claude Code broadly, gives teams uncapped AI spend, and extends AI adoption beyond engineering into finance, sales, and support. The company pairs with AI labs, builds internal MCP and LLM infrastructure, and even incorporates AI into hiring and management workflows.
Main Topics: Shopify’s AI-first engineering culture: Shopify treats AI as a default capability across the company, not just a tooling choice for engineers. Farhan frames the organization as problem-solvers who pair on issues rather than staying inside rigid role boundaries. Early adoption and experimentation with AI coding tools: The company was an early user of GitHub Copilot, later adopted Cursor, and now tests Claude Code, Devon, and other tools. The approach is to try tools, measure what works, and expand usage carefully. Internal AI infrastructure: LLM proxy and MCP: Shopify built internal infrastructure to secure model access, control costs, and expose company systems through MCP endpoints. This lets employees query internal knowledge and connect models to tools and data safely. AI beyond engineering: business users building software: Cursor and related tools are being used heavily by finance, sales, support, and other non-R&D teams. These teams are building personal workflows and MCP-based automations without heavy engineer involvement. Cost, productivity, and the AI spend philosophy: Shopify argues AI spend should be treated as investment, not line-item optimization. Leadership believes token and model costs are justified if they improve productivity, even if exact ROI is hard to measure. AI in hiring, interns, and management workflows: Shopify uses AI in interviews for senior hires, expects reflexive AI use internally, and is expanding its internship program to bring in AI-native talent. AI also helps draft weekly project updates and surface team metrics. Limits and guardrails: human judgment still matters: Despite enthusiasm for AI, Shopify emphasizes understanding code, reviewing AI output, and avoiding blindly accepting generated work. The team sees the current era as human-in-the-loop and not yet ready to replace foundational understanding.
Key Arguments: Shopify pairs with AI labs and vendors rather than just consuming products, because close collaboration helps shape tools and learn faster. AI adoption should be broad and expected; employees may not be forced to use every tool, but their output is judged as if they had access to them. Non-engineering teams can and should build useful software for themselves, especially for personal workflows and automation. The current AI era is best understood as a middle zone: useful for prototypes, personal tools, and productivity, but not yet reliable enough to fully build internet-scale infrastructure without expert oversight. High AI spend can be rational if it delivers meaningful productivity gains; leadership should not be penny-pinching on models and tokens. The best engineers will use AI to attack technical debt, refactoring, and infrastructure cleanup that they previously lacked time for. Interns are seen as a source of learning and cultural change, especially because younger workers are more AI-reflexive. Role modeling matters most: leaders must visibly use AI tools, share prompts, and demonstrate workflows to drive adoption.
Data Points: Shopify engineers: 3,000 - Approximate company engineering headcount discussed during the interview Interns hired per term last year: 25 - Previous scale of Shopify’s internship program before expansion Interns planned for this year: 1,000 - Target internship hiring over the year, driven by AI-reflexive talent strategy Interns per term at new scale: 350 - Rough term-by-term breakdown of the 1,000 intern plan Company Summit attendance: 7,000 people - Employee event used as an example of hands-on leadership, including Wi‑Fi troubleshooting Ubiquiti access points deployed: 300 APs - Infrastructure deployed for improved Summit Wi‑Fi Switches deployed: 800 switches - Additional networking hardware deployed for Summit Wi‑Fi Code red duration: Almost 7 months - Engineering-wide focus on reliability and tech-debt reduction Code red initial expectation: 3 months - Original estimate before the reliability initiative stretched longer Engineering focus during code red: 30% to 50% - Approximate portion of engineering dedicated to reliability work 28-day rolling average: Used as a metric - Reliability and operational stability were tracked using rolling averages AI tool pricing mentioned: $10-$20/month - Reference to older Copilot/ChatGPT-style pricing comparisons Higher-end AI pricing mentioned: $200/month - Premium model subscriptions leaders urged people to consider Extreme AI spend example: $10,000/month per engineer - A reported case Farhan wanted to understand because it seemed unusually high ChatGPT launch timing: 2022 - Referenced as the period when Shopify’s AI infrastructure work accelerated Copilot adoption timing: 2021 - Shopify requested Copilot before ChatGPT existed Cursor adoption timeline: About 1 year - Farhan said Shopify was relatively new to Cursor MCP servers at Shopify: Two dozen - Internal endpoints already available and growing quickly LLM usage expectation: Every employee expected to have tools available - Internal policy shift described as moving from passive to active AI use
Pivotal Quotes: "hire smart people and pair with them on problems" — Farhan Tavar: Explaining Shopify’s management style and why leaders stay hands-on "you don't have to use AI in your team, in your workflow, but we're going to expect that you have this plethora of tools available to you" — Farhan Tavar: Describing Shopify’s internal AI policy and expectation-setting memo "we're very much in this middle zone right now" — Farhan Tavar: Summarizing the current limits of vibe coding and AI-built software
Implications: Shopify is treating AI as core operating infrastructure, not a side experiment. The model suggests future winners will pair human judgment with AI tools, expand access beyond engineering, and build internal platforms that make AI safe, searchable, and measurable.