Agents in the Wild

Shopify's River agent coauthored about one in eight merged pull requests from inside public Slack

River is Shopify's internal AI agent. Employees mention it in public Slack channels, where it reads code, runs tests, opens pull requests, queries data and checks production traces. In 30 days it ran 59,918 sessions.

Organization
Shopify
Industry
E-commerce software
When
May 2026
Evidence
Company reported
Agents and tools
RiverAquiferSlackNix

Reported outcome

Shopify reports River coauthored about one in eight merged pull requests, and 3,536 merged pull requests in one 30-day window.

  • about one in eight (company)
    Merged PRs coauthored
  • 59,918 (company)
    Sessions in 30 days
  • 5,170 (company)
    Slack channels touched
  • 7,000+ (company)
    People involved
  • 19 minutes, 50 tool calls (company)
    Median session

Caveats

Stated by the company that uses the agent (blog, earnings call, interview). Not independently audited.

  • All numbers are Shopify's own, from a table River writes about itself.
  • Counts of sessions and merged pull requests do not show code quality or business value.
  • The post does not name the AI models and does not describe a formal review step beyond normal merging.
  • The authors say the numbers were already out of date when published.

What the agent does

River lives in Shopify's Slack. Employees @-mention it in public channels. It reads code, runs tests, opens pull requests, queries the data warehouse and looks at production traces. Sometimes it pushes back on a plan it thinks is flawed.

River works only in public channels, never in direct messages. So every conversation is a transcript that other employees can find and learn from. People join threads to add limits or change direction.

Why this works

River runs on Aquifer, Shopify's internal platform for agents (sessions, sandbox, gateway, credentials proxy, event log, observability). It relies on earlier bets: one monorepo called World and Nix for reproducible environments. Skills, AGENTS.md files and runbooks live in the repo.

What to copy

  • Put the agent where work already happens, and make its threads public.
  • Give it a reproducible environment so it can run tests itself.
  • Log every session to a table so you can measure use.

Starter prompt

Written by us, not by Shopify.

You are our team's helper in a public Slack channel.
Request: {request}
Repo and tools you can use: {tools}

Before changing code, reply with a short plan and wait for a thumbs-up.
Run the tests. Open a pull request with a clear summary.
If you think the request is a bad idea, say why in one or two sentences.

Sources

  1. Under the River — Shopify Engineering, May 28, 2026

Added Nov 2, 2026 · Sources last checked Nov 2, 2026. Spotted an error? Send a correction.

Get new use cases weekly

One short email. Sources included.

Related use cases