How to Make AI Work Visible at Scale: Lessons from Shopify's River Agent
Shopify's River agent runs only in public Slack channels. Here's how to apply the same principle to build shared AI taste across your organization.
The Problem with AI That Nobody Can See
Most organizations deploy AI in the worst possible way for long-term adoption: in private.
Someone uses a ChatGPT window on their laptop. Another person asks Claude for a draft in a personal account. A third builds a quiet automation that runs in the background. All of this is useful to the individual, but it’s invisible to everyone else. The organization never learns. Norms never form. And the gap between people who are quietly compounding their AI skills and those who aren’t grows wider every month.
Shopify saw this problem and built something specifically to solve it. Their internal AI agent, River, operates under a single defining constraint: it only runs in public Slack channels. No DMs. No private groups. Every conversation with River happens where the rest of the company can see it.
That one rule changes everything about how AI spreads through an organization. This article breaks down why visibility matters for enterprise AI, what lessons River holds for companies trying to build shared capability at scale, and how to apply these principles whether you’re deploying a multi-agent system or building your first automated workflow.
Why Visibility Is the Missing Variable in Enterprise AI
When you ask why some companies are pulling real value from AI while others are stuck in pilot purgatory, the answer usually isn’t about which model they’re using or how much budget they’ve allocated. It’s about whether AI use is visible and social within the organization.
Remy doesn't build the plumbing. It inherits it.
Other agents wire up auth, databases, models, and integrations from scratch every time you ask them to build something.
Remy ships with all of it from MindStudio — so every cycle goes into the app you actually want.
Visible AI work does a few things that private AI use simply can’t:
- It normalizes AI assistance. When people see colleagues using AI in Slack, it stops feeling like cheating or cutting corners. It becomes what professionals do.
- It teaches through observation. Watching a skilled colleague prompt an AI agent teaches you more in 30 seconds than a training document would in an hour.
- It surfaces standards. Organizations can’t develop shared quality norms for AI outputs if no one can see what good looks like.
- It creates accountability. AI suggestions that appear in public channels get scrutinized. That’s a feature, not a bug — it keeps quality high and prevents over-reliance on bad outputs.
Shopify’s River was designed with this logic baked in from the start. It’s not accidental that the agent works only in public channels. That constraint is the product.
What Shopify’s River Agent Actually Does
River is Shopify’s internal AI coding and knowledge assistant. It’s deeply integrated into how engineering teams work — answering questions about codebases, helping with architecture decisions, summarizing context, and assisting with technical tasks.
But the design philosophy behind River is what makes it interesting beyond the technical spec.
The Public Channel Rule
Shopify made a deliberate architectural choice: River doesn’t operate in DMs or private Slack channels. If you want help from River, you ask in a public channel. This means every interaction — every question asked, every answer given, every correction made — is visible to anyone in the organization who cares to look.
This sounds simple. It is simple. And it has significant downstream effects.
Building Shared AI Taste
One concept Shopify has talked about internally is building “taste” for AI — a shared organizational sense of what good AI use looks like, what prompts work well, which outputs need scrutiny, and how to iterate.
Taste is almost impossible to build in private. It requires social transmission. You develop it by watching others, by seeing what gets praised and what gets corrected, by absorbing patterns over time. When River operates in public, all of that learning happens passively and continuously.
Reducing the Knowledge Gap
One of the most corrosive dynamics in AI adoption is the emergence of a two-tier workforce: the people who figured out how to use AI well, and everyone else. This gap isn’t about intelligence — it’s about exposure and feedback loops.
Public AI use breaks down this dynamic. When a strong engineer asks River a complex question in a Slack channel, everyone who reads that thread gets a free lesson. The learning distributes naturally.
The Principles Behind Making AI Work Visible
River’s design isn’t just a quirky rule. It reflects a set of principles that any organization can apply when thinking about how to deploy AI at scale.
Principle 1: Default to Public, Opt Into Private
One coffee. One working app.
You bring the idea. Remy manages the project.
Most enterprise tools default to private. Slack messages are private unless you’re in a public channel. Emails are private. Document drafts are private. This default toward privacy makes sense for sensitive work — but it’s the wrong default for AI-assisted work that involves broadly applicable knowledge.
The question to ask when configuring an AI tool or agent in your organization isn’t “should this be private?” It should be “does this need to be private?” Customer data, HR matters, and competitive strategy need to be private. Code questions, writing assistance, research tasks, and analytical work usually don’t.
Flip the default.
Principle 2: Make the Agent a Team Member, Not a Private Tool
River shows up in Slack as a participant. It’s not a sidebar widget or a separate app. This framing matters. When AI is positioned as a team tool that’s part of the working environment, it gets used in the flow of work — not as a side activity that people do quietly on their own time.
This is also why the channel structure matters. An agent that lives in #engineering-help gets invoked when people have engineering questions. An agent that lives only in a private interface gets used only by the people who already know to use it.
Principle 3: Treat AI Outputs as Discussion Inputs, Not Final Answers
One risk with invisible AI work is that people treat AI outputs as endpoints — they get an answer and act on it without socializing or scrutinizing it. Public channels naturally create a review layer. When River responds to a question in Slack, other engineers can weigh in, push back, or add context. The AI output becomes a starting point for discussion, not the end of a private conversation.
This is healthier epistemically. It prevents over-reliance and keeps human judgment in the loop.
Principle 4: Let Patterns Accumulate Publicly
One underrated benefit of public AI use is that the historical record becomes an organizational asset. When every question asked to River and every response given is visible in a shared Slack channel, that history is searchable. New employees can browse it. Teams can review patterns over time. Managers can see where people are getting stuck and where the AI is doing well.
This kind of ambient record-keeping is impossible when AI use is scattered across private windows and personal accounts.
How to Apply This at Scale: A Practical Framework
You don’t need to be Shopify to apply River’s design principles. Here’s how to think about rolling this out in an organization of any size.
Step 1: Identify the Right Surface
The first question is where your AI agent should live. Slack and Microsoft Teams are the obvious choices for most organizations because they’re already where work happens. But the principle applies anywhere — it could be a shared project management tool, a centralized knowledge base, or a ticketing system.
The key is that the surface needs to be:
- Accessible by default to the relevant team or organization
- Persistent, so the history of AI interactions stays searchable
- Social, meaning there’s a natural mechanism for others to read, react, and respond
Step 2: Set Channel Structure Intentionally
Seven tools to build an app. Or just Remy.
Editor, preview, AI agents, deploy — all in one tab. Nothing to install.
Don’t just drop an AI agent into a general channel and call it done. Think through the channel structure based on use cases:
#ai-research-helpfor knowledge lookup and synthesis tasks#ai-code-reviewfor engineering teams#ai-writing-assistfor marketing and content teams#ai-analysisfor data and operations work
This segmentation serves a few purposes. It makes outputs searchable by type. It allows you to configure the agent differently for different contexts. And it helps people know where to go when they need a specific kind of help.
Step 3: Configure the Agent with Organizational Context
A generic AI assistant deployed into Slack is marginally useful. An AI agent configured with your organization’s specific context — your codebase, your brand guidelines, your internal knowledge base, your data sources — is substantially more useful.
River works well in part because it has deep access to Shopify’s internal systems. Your version of River should have access to the information that makes it relevant to your team’s actual work.
Step 4: Establish Norms Early
The first few weeks of a public AI deployment are critical for norm formation. When the agent gives a bad answer, someone should correct it publicly. When someone asks an unusually good question that elicits a useful response, acknowledge it.
Consider pinning a brief document in the channel that covers:
- What the agent is designed to help with
- What it’s not good at
- How to prompt it effectively for your use case
- How to flag errors or quality issues
Step 5: Review Usage Patterns Periodically
Every 30–60 days, review what kinds of questions are flowing through your AI channels. This review should cover:
- What topics come up most frequently
- Where the AI is consistently helpful vs. where it struggles
- Whether certain teams or individuals are using it much more (or less) than others
- Whether the outputs have stayed consistent in quality
This feedback loop is what lets you improve the agent over time — something that’s impossible to do when usage is scattered and invisible.
Multi-Agent Visibility: The Next Layer
As organizations mature in their AI use, the challenge shifts from getting people to use AI to coordinating multiple AI agents that work together across different parts of the organization.
This is where the visibility principle becomes even more important — and more complex.
Making Agent-to-Agent Work Visible
In a multi-agent workflow, one agent might research a topic, pass results to another agent that synthesizes them, and a third agent that formats the output for a specific audience. All of this can happen invisibly in the background.
The question is: how much of this should surface in human-visible channels?
The answer depends on the stakes of the task. For high-stakes workflows — anything touching customer data, financial outputs, or public communications — you want checkpoints where human-readable summaries post to a shared channel before the workflow proceeds. For low-stakes routine tasks, full automation without visibility is fine.
The design principle: higher stakes require more visibility, not less.
Audit Trails as Organizational Infrastructure
How Remy works. You talk. Remy ships.
One of the most practical arguments for visible AI work at scale is the audit trail. When AI-assisted decisions are made in visible, logged channels, there’s a record of what information the AI had access to, what it recommended, and what the human decided to do. This matters for compliance, for troubleshooting, and for continuous improvement.
Organizations that let AI work happen invisibly are essentially operating without logs. That’s fine until something goes wrong — and then it’s a serious problem.
How MindStudio Fits Into This Picture
The River principle is compelling, but most organizations don’t have Shopify’s engineering resources to build a custom internal agent from scratch. This is where MindStudio becomes relevant.
MindStudio is a no-code platform for building and deploying AI agents. It connects to Slack natively — you can build an agent that responds to messages in specific channels, posts outputs to designated workspaces, and maintains visibility into everything it does without requiring any backend infrastructure.
Where this gets practically useful:
- You can build a River-style agent specific to your organization — with access to your knowledge bases, your Notion docs, your Airtable data, your internal tools — in a few hours rather than months.
- MindStudio’s Slack integration means your agent can live directly in public channels, not in a separate interface that people have to remember to visit.
- Because MindStudio supports multi-agent workflows, you can chain agents — one that handles research, one that drafts, one that formats — and configure which steps post updates to visible Slack channels and which run silently in the background.
The point isn’t that MindStudio is the only way to build this. It’s that the Shopify design principle — AI work should be visible, social, and public by default — can be implemented without a large engineering team if you have the right platform underneath it.
You can try MindStudio free at mindstudio.ai.
FAQ
What is Shopify’s River agent?
River is Shopify’s internal AI agent, primarily used to help engineering teams with coding questions, codebase navigation, and technical knowledge. It’s integrated into Slack and operates under a specific design constraint: it only responds in public channels, not in private messages or closed groups. This constraint was intentional — it ensures that AI-assisted work is visible across the organization.
Why does River only work in public Slack channels?
The decision to restrict River to public channels was made to make AI use visible and social. When AI interactions happen in public, they normalize AI use, create shared learning opportunities, surface organizational norms about quality, and build a searchable history of how AI has been used. Private AI use creates none of these benefits.
How do you build organizational AI culture at scale?
Building AI culture at scale requires three things: making AI use visible (as Shopify does with River), creating feedback mechanisms so people can observe and learn from each other, and establishing shared standards for what good AI output looks like. Visibility is the prerequisite — culture can’t form around practices that nobody can see.
What are the risks of letting AI work happen in private?
Plans first. Then code.
Remy writes the spec, manages the build, and ships the app.
When AI use is private and scattered across personal accounts and individual tools, organizations lose the ability to set quality standards, share learning, build consistent practices, or maintain audit trails. It also creates an uneven playing field where some employees quietly compound their AI skills while others fall behind. Over time, this fragments the workforce and makes organizational AI strategy nearly impossible to execute.
How do you scale AI agents across a large organization?
Scaling AI agents effectively requires a deliberate channel structure (so different agents serve different functions in the right contexts), organizational context baked into the agent (not just a generic model), clear norms about when to override or question AI outputs, and periodic review of usage patterns. The structural piece — putting agents where work already happens rather than requiring people to seek them out — is consistently the most important factor for adoption.
What’s the difference between a private AI tool and a public AI agent?
A private AI tool is one that individuals use on their own, with outputs and interactions visible only to them. A public AI agent is one deployed in a shared environment where interactions are visible to the team or organization. The distinction matters because public agents create social learning, accountability, and accumulated organizational knowledge. Private tools create individual productivity gains but don’t build collective capability.
Key Takeaways
- Shopify’s River agent operates only in public Slack channels — a deliberate design choice that makes AI use visible, social, and learnable across the organization.
- Invisible AI work creates an uneven workforce, prevents norm formation, and eliminates the possibility of building shared organizational capability.
- The core principle is simple: default to public for AI interactions, opt into private only when there’s a specific reason to do so.
- Making AI work visible requires intentional channel structure, agent configuration with organizational context, and periodic review of usage patterns.
- Multi-agent workflows require explicit decisions about which steps surface to humans and which run silently — higher-stakes tasks should have more visibility checkpoints, not fewer.
- Platforms like MindStudio let organizations build River-style public AI agents without deep engineering investment, using native Slack integration and no-code workflow building.
The organizations that get ahead with AI over the next few years won’t just be the ones with access to the best models. They’ll be the ones that figured out how to make AI capability a shared, social, and compounding organizational asset — not a private perk for whoever happened to find a good ChatGPT prompt.
