Case Study: Shopify River, the agent platform that rebuilt the reference architecture
One in eight pull requests merged across Shopify is coauthored by River, its internal Slack-native coding agent. The number is not the lesson. The lesson is the platform underneath: River is one profile on Aquifer, "Shopify's internal platform for running AI agents," and Aquifer's architecture, derived independently, is the same brain/hands/durable-session shape this guide calls the reference architecture. Among the eight case studies here, River is the strongest public evidence that the architecture is convergent, not one vendor's opinion.
Aquifer's governing constraint is that "the session must survive" process deaths and machine failures. Because "the work lives in Postgres, not in memory," a session outlives the process running it and the host it ran on. That is exactly the session this guide argues for: the resume mechanism and the audit record are the same artifact, so durability and auditability come from one design, not two.
Aquifer separates the brain from the hands as two tiers. "The harness lives outside the sandbox. The agent doesn't live where the code lives," so "the agent loop is not in the same blast radius as rm -rf." This is Anthropic's brain/hands split and the one-environment-per-task discipline reached without reference to either: the same constraints produced the same design.