[1/2] How AI Is Changing Software Engineering at Shopify (Category AI)
I watched an interesting interview with Gergely Orosz and Farhan Thawar, who oversees more than 3 000 engineers at Shopify. Before Shopify, Thawar worked at Pivotal Labs and Xtreme Labs and co-founded Helpful.com, which Shopify acquired. He is known for hands-on leadership: in the interview he even describes personally fixing Wi-Fi at a company event :). Orosz writes The Pragmatic Engineer, publishes books and speaks at conferences. I recently discussed his “Software Engineering with LLMs in 2025: Reality Check” talk.
Here are the interview’s main ideas:
1. Shopify’s philosophy: pair on problems Rather than “hire smart people and leave them alone,” Shopify’s approach is to hire smart people and work alongside them on problems. Thawar says everyone is expected to solve problems regardless of title; leaders cannot simply delegate them downwards.
2. Early AI tool adoption Shopify was an early GitHub Copilot user. In 2021, before ChatGPT launched, Thawar personally asked GitHub’s new CEO for access for all Shopify engineers. The company received two years of free access in exchange for feedback.
3. Diversifying AI tools Shopify traditionally preferred one tool per task, but changed that approach for AI to avoid putting all its eggs in one basket:
- GitHub Copilot and Cursor for development.
- Claude Code for agent workflows.
- Devin, which the company tested.
4. AI beyond engineering Finance, sales and support teams also use Cursor actively. Nontechnical employees build MCP servers to access Salesforce, Google Calendar, Gmail and Slack, often without engineers’ help.
5. Code Red: tackling technical debt Shopify ran a 7-month Code Red programme from November 2024 to June 2025, assigning 30–50% of engineers to technical debt.
6. Shopify’s CEO memo on AI Tobi Lütke’s internal memo made AI use mandatory for employees:
- Reflexive AI use is a baseline expectation.
- Before requesting more resources, people must demonstrate that AI cannot do the job.
- AI competence becomes part of performance assessment.
7. AI infrastructure
- An internal LLM proxy supports secure model use without exposing company data, tracks tokens and includes a leaderboard of the highest-consuming users.
- More than two dozen Model Context Protocol servers provide internal data access, including The Vault wiki with project information and presentation transcripts.
8. The philosophy of AI spending Shopify does not cap AI token spending. Thawar argues that even $1000 per engineer per month is “too cheap” for tools delivering a 10% productivity gain. The company encourages advanced rather than basic models.
9. Internships Shopify plans to hire 1000 interns in 2025, calling them “AI centaurs”: people who work effectively with language models. The internships last 4 months and require office attendance in Toronto, Ottawa or Montreal to build a culture of learning together.
My conclusions are in the next part.
#AI #ML #PlatformEngineering #Software #Architecture #Processes #DevEx #Devops