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

Page: https://stenobird.com/podcast/latent-space-ai-engineer/shopify-s-ai-phase-transition-2026-usage-explosion-unlimited-opus-4-6-token-budget-tangle-tangent-simgym-with-mikhail-parakhin-shopify-cto
Text version: https://stenobird.com/podcast/latent-space-ai-engineer/shopify-s-ai-phase-transition-2026-usage-explosion-unlimited-opus-4-6-token-budget-tangle-tangent-simgym-with-mikhail-parakhin-shopify-cto.md
Podcast: [Latent Space: The AI Engineer Podcast](https://stenobird.com/podcast/latent-space-ai-engineer)
Published: 2026-04-22T19:33:00+00:00
Episode link: https://www.latent.space/p/shopify
Audio file: https://api.substack.com/feed/podcast/195067855/708627edff5e672fd6cb54ad90b94b0c.mp3
Processing state: processed
JSON: https://stenobird.com/v1/public/podcasts/latent-space-ai-engineer/episodes/shopify-s-ai-phase-transition-2026-usage-explosion-unlimited-opus-4-6-token-budget-tangle-tangent-simgym-with-mikhail-parakhin-shopify-cto
Duration seconds: 4345

## Resource

Shopify CTO Mikhail Parakhin reveals how the company is moving beyond simple AI adoption to building a proprietary ecosystem of simulation and optimization tools. The discussion explores the shift from code generation to the much harder problems of automated review, deployment stability, and customer behavior simulation.

## Highlights
- Main idea: The real bottleneck in the AI era has shifted from code generation to the complexity of review, CI/CO, and deployment stability
- Practical takeaway: Effective AI engineering requires investing more in critique loops and automated testing than in raw token generation
- Failure mode: Relying solely on high token counts is a poor metric for engineering output; the focus must be on the quality of the feedback loop
- Main idea: Shopify's 'SimGym' uses large-scale agentic simulations to model complex human and company counterfactuals
- Technical insight: Internal tools like Tangle use content hashing to make massive ML and data workflows reproducible and efficient

## Topics

AI Engineering, Machine Learning Operations, Customer Simulation, Automated Software Engineering, Large Language Models, Software Infrastructure, Agentic Workflows, Data Reproducibility

## Chapters
- 1:00 — The Shift to Internal AI Adoption: Mikhail discusses the recent surge in AI tool adoption within Shopify and the rise of CLI-based development tools.
- 6:40 — The Token Budget Fallacy: A debate on whether raw token count is a meaningful metric for evaluating the success of AI-driven engineering.
- 12:00 — Agentic Reasoning vs. Parallelism: Why high-quality, slow-turn models are more effective for complex tasks than large swarms of parallel agents.
- 23:00 — Efficiency through Content Hashing: How Shopify uses hashing in Tangle to ensure data preprocessing and experiments are only rerun when necessary.
- 34:05 — The Reality of Automated Research: Reflections on the low success rate of automated experiments and the value of automated optimization.
- 50:15 — Simulating Complex Ecosystems: An introduction to SimGym and the power of modeling counterfactuals in complex business environments.
- 55:50 — The Future of State Space Models: A technical look at the limitations of current architectures and the potential of liquid neural networks.

## Actions

- request_transcript: `POST https://stenobird.com/v1/public/podcasts/latent-space-ai-engineer/episodes/shopify-s-ai-phase-transition-2026-usage-explosion-unlimited-opus-4-6-token-budget-tangle-tangent-simgym-with-mikhail-parakhin-shopify-cto/transcription-requests` — Idempotently request low-priority transcript generation for this episode.
- read_markdown: `GET https://stenobird.com/podcast/latent-space-ai-engineer/shopify-s-ai-phase-transition-2026-usage-explosion-unlimited-opus-4-6-token-budget-tangle-tangent-simgym-with-mikhail-parakhin-shopify-cto.md` — Read the agent-friendly Markdown representation of this episode resource.

A page view does not enqueue transcription. Agents should invoke `request_transcript` explicitly when they need this episode processed.

## Transcript

Full transcripts are not published on public pages unless there is a clear rights basis.