Shopify’s AI Phase Transition: 2026 Usage Explosion, Unlimited Opus-4.6 Token Budget, Tangle, Tangent, SimGym — with Mikhail Parakhin, Shopify CTO

This section explores how Shopify has evolved into a leader in AI integration across its internal systems and tools. Mikhail Parakhin discusses the company’s recent push towards AI, including recent efforts like internal tooling, Tangle, and adoption metrics. The rapid growth of AI tool usage among Shopify employees, especially after a December inflection point where AI models significantly improved, highlights Shopify's aggressive and natural progression in AI adoption. Parakhin emphasizes the importance of AI in boosting productivity, optimizing workflows, and the company's goal to be at the technological frontier.

Parakhin presents Shopify’s unique systems for ML experimentation—Tangle and Tangent—that enable reproducible, collaborative, and efficient data workflows. Tangle acts as a third-generation data processing and ML experimentation engine that simplifies operations like feature engineering, experiment management, and rollback. Tangent enhances this ecosystem by automating auto-research and hyperparameter optimization through auto-iteration, allowing even product managers and domain experts to run experiments and optimize without deep ML expertise. These innovations dramatically accelerate R&D and improve decision-making.

Shopify leverages advanced customer simulation models, particularly through their SimGym platform, which replicates merchant and customer behaviors based on historical data. The system uses AI agents to simulate real shop interactions, enabling merchants to predict outcomes of changes and optimize conversions. Parakhin discusses how these simulations, especially with visual and multi-modal data, are computationally expensive but powerful for improving marketing strategies and operational decisions. The platform’s ability to model complex trajectories and counterfactual scenarios offers a huge competitive moat, especially for large-scale Shopify merchants.

Parakhin explains Shopify’s exploration of alternative AI architectures, particularly Liquid AI, which is inspired by neural models based on differential equations rather than transformers. Liquid AI is highly efficient, especially for long-context tasks with low latency requirements, such as search and categorization. Shopify actively deploys Liquid models internally, finding them competitive for specific use cases. The discussion also touches on the potential for large-scale Liquid models if more compute were available, and the interest in hybrid models combining transformers and Liquid. Parakhin emphasizes the importance of practical, scalable AI architectures tailored to different tasks.

This section covers the technical hurdles of serving large models and managing extensive AI workloads at scale. Parakhin talks about innovative infrastructure solutions, including custom model tuning with Santa ML and NVIDIA, browser-based serving, and content-addressed caching to optimize costs and efficiency. He discusses how current pipelines like GitHub's PR flow face bottlenecks under AI-driven development and suggests that traditional code and deployment models may need radical redesigns, possibly adopting microservices or new paradigms as AI development accelerates.

Parakhin discusses Shopify’s ongoing hiring efforts across ML, data science, and distributed systems to fuel further innovation. He highlights the company’s active engagement in cutting-edge projects, including using LLMs for reimagining distributed databases and AI-driven experimentation. He underscores Shopify’s commitment to keeping at the forefront of AI, constantly testing new models and architectures, including multi-modal and hybrid systems. The emphasis is on attracting talent that can push the boundaries of scalable, practical AI applications.

Looking at the history of Shopify’s AI, Parakhin shares insights into the early days of the Sydney chatbot at Bing, including deliberate personality shaping to make interactions more engaging and human-like. He reveals how Shopify’s internal work continues in creating AI personas and how these experiments aim to improve user experience. The discussion touches on the potential for future AI personalities and deeper contextual models to transform customer engagement and operations, reinforcing Shopify’s unique position in leveraging AI for enterprise-level impact.