Under the River

Summary

Shopify describes how its AI agent, River, grew from an infrastructure bet aimed at making a large, fragmented codebase usable by AI. In early 2024, the company adopted a monorepo called World and Nix as a reproducible substrate for development, CI, and production, despite substantial migration costs. River operates in public Slack channels, where it reads code, runs tests, queries data, inspects production traces, and opens pull requests; in a recent 30-day period it handled 59,918 sessions and contributed to 3,536 merged pull requests. Underneath it, Aquifer provides durable sessions, sandboxes, gateways, event logs, credentials handling, and observability, allowing agent profiles to share one platform across interactive, automated, and batch modes. The central design principle is that the session must survive process and machine failure, while public transcripts create a compounding knowledge corpus.

Context

Shopify had many repositories, bespoke development environments, and slow feedback. The company also identified a ceiling in local, private coding agents: knowledge from one engineer’s agent session did not automatically benefit other people or future sessions. As more teams requested specialized agents, Shopify needed a shared foundation rather than separate platforms.

Approach / What changed

Shopify consolidated its code into the World monorepo and built development, CI, and production environments with Nix. River was implemented as a public Slack-based agent, while Aquifer became the underlying platform for durable sessions, harnesses, sandboxes, gateways, event logs, credential proxies, and observability. Agent products are represented as profiles containing prompts, skills, extensions, sandbox policies, and model defaults, and the same substrate supports interactive, automated, and batch workloads.

Takeaways

• River is restricted to public Slack channels, making its conversations searchable and allowing fixes, constraints, and discoveries to become reusable knowledge for other employees and future sessions.
• Aquifer separates an agent’s decision-making brain from the sandbox where tools run, while preserving the conversation across the failure or replacement of cells, sandboxes, and machines.
• The platform uses one session model, sandbox plane, and gateway for interactive River sessions, automated pull-request review, and ephemeral CI or batch jobs.

Related reading