Shopify's River Won't Talk to You in Private — That's the Point
Agentic AIShopify’s internal coding agent, River, does not respond to direct messages. If you want to work with it, you create a public Slack channel. Every conversation is searchable by everyone in the organization. That design decision — forcing transparency at the agent interaction layer — is more interesting than anything River actually does.
Why Public-Only Is a Structural Choice, Not a Feature
Most enterprise AI deployments optimize for individual productivity: a developer asks a coding agent a question in a private context, gets an answer, moves on. Knowledge about how to prompt effectively, what the agent can and can’t do, and what workflows produce good results stays with that individual. It does not accumulate.
Shopify CEO Tobias Lütke frames River’s design around a concept he calls the Lehrwerkstatt — the German teaching workshop model where apprentices learn by doing visible work alongside experienced practitioners. River forces the same dynamic: because every agent session runs in a public channel, every session is an observable learning event for anyone watching. Lütke’s own #tobi_river channel has over 100 participants who react to threads, offer context, contribute to code reviews, and — most importantly — learn how a senior practitioner interacts with an AI coding agent by watching it happen in real time.
Simon Willison draws a parallel to Midjourney’s early architecture: public Discord channels as the primary interaction surface created a shared knowledge base that no onboarding document could replicate. Users learned what good prompts looked like by seeing thousands of examples in context. River applies the same mechanic to enterprise software development.
What This Means for How Organizations Deploy Agentic AI
Most enterprise agentic AI deployment frameworks treat access control as the primary architectural consideration: who can use the agent, what data it can touch, what actions it can take. Shopify’s approach adds a second axis: what’s the ambient observability model for agent usage across the organization?
Private-by-default deployment optimizes for security and individual comfort. Public-by-default deployment optimizes for organizational learning velocity. Neither is universally correct — sensitive contexts (HR, legal, compensation) clearly require private sessions. But for engineering work, where the quality of AI interaction patterns matters to team productivity across hundreds of developers, the public model may produce compounding returns that private deployment cannot.
The Claude Code channels architecture covers related terrain from a tooling perspective: persistent agent sessions as team infrastructure rather than individual utilities. River takes that architectural idea and adds organizational transparency as a forcing function.
The So What
River is a coding agent with an unusual constraint: it cannot be used quietly. The constraint is the product. Shopify is betting that the value of shared, observable AI interaction patterns — accumulated across a public channel history that any engineer can search — outweighs the friction of working in the open. For enterprise platform teams trying to scale agentic AI adoption beyond early adopters, that bet is worth examining. The question is not whether your agents are capable. It is whether the organization is learning faster because of how they’re deployed.
Photo by Thirdman on Pexels
Source: Simon Willison's Weblog
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →