Building reliable and proactive agentic systems at scale: how Shopify’s reflexive AI culture was instrumental in their development of Sidekick w/ Andrew McNamara #258

At Shopify, 95% of employees now use AI daily, creating a 'reflexive' culture where AI is embedded in every stage of product development—from prototyping to deployment. This cultural shift, powered by a commitment to hiring early talent and building at the lowest possible level, enabled the creation of Sidekick, an AI co-founder that acts as a business mentor for millions of merchants. The key to its success lies not in complex multi-agent architectures, but in a simple orchestration model with specialized sub-agents, like a fine-tuned GraphQL tool that outperforms Opus 4.7. Central to this system is a constantly evolving 'ground truth set' of human-rated conversations, which serves as the foundation for AI judges that enable auto-research, prompt optimization, and scalable evaluation. Sidekick Pulse, the proactive variant, uses background research and high-reasoning loops to deliver actionable insights—like fixing US shipping misconfigurations—without user input. Reliability is ensured through model-agnostic, cloud-agnostic redundancy and a strict 'show intelligence within a second' principle that balances deep thinking with responsiveness. The entire platform is built on one core mantra: keep building.