← All implementations

Coinbase · Agent system

Forge / Mux

Forge, started as Claudebot, turns a Slack/GitHub/Linear discussion into a Linear issue, fix, PR, and one-off build; Mux lets employees run many coding agents concurrently; as of February 2026, 5% of all PRs merged company-wide come from these background agents.

1 Supports2 Supports3 Supports5 Supports

Approach type
Agent system
Work
Coding, Code review
Human involvement
Drafts reviewed
Invocation
Event driven, Interactive
Interfaces
Slack, Github, Linear
Deployment stage
Scaled
Evidence strength
Detailed primary
Entry reviewed

How it works

The workflow the sources report for this implementation.

Custom harness: Slack bug discussion → Linear issue → fix → PR → one-off build

2 Supports5 Supports

Concurrency layer; one human coordinates many isolated coding agents, each in its own worktree/branch/terminal

1

Where people stay involved

  • Slack, GitHub, or Linear request → reviewed pull request and buildWork product review · Level 3

Level 3 for Slack, GitHub, or Linear request → reviewed pull request and build; human attention boundary: work-product-review.

1 Supports3 Supports

Observation date
2026

Implementation details

Sandbox

Mux gives each concurrent agent its own git worktree, branch, and terminal

1

Harness

Portfolio approach; Claude Code, OpenCode, Cursor, and Copilot rather than one harness; Forge is a custom Slack-native harness with the same tools and context as engineers, invokable from Slack, GitHub, and Linear; built in-house because security requirements blocked cloud background agents

1 Supports2 Supports3 Supports5 Supports

Model

Portfolio across multiple providers; the Slack agent Claude bot can use multiple underlying models despite its name

1 Supports5 Supports

Interfaces

slack, github, linear

1

Knowledge

Linear as the durable structured-context layer for product work

2

Reported results and limitations

The catalog records what the sources report, with the scope and the denominator of every figure. A qualification below limits the figure it sits under.

Reported metrics

Headline claim

5% of all PRs merged come from in-house background agents (Feb 2026)

3

Reported by
Coinbase
Scope
PRs merged by in-house background agents
Denominator
All PRs merged company-wide
Observation date
2026-02
Key observation

Mux: 600+ users including engineers, PMs, and designers (335 active, 197 power users)

1

The source does not report the denominator of this figure.

Reported by
Coinbase
Scope
Registered Mux users including engineers, PMs, and designers; 335 active and 197 power users
Key observation

5% of all PRs merged come from in-house background agents (Feb 2026)

3

Reported by
Coinbase
Scope
PRs merged by in-house background agents
Denominator
All PRs merged company-wide
Observation date
2026-02
Key observation

Wallet rewrite PR review times reduced from 150 hours to 15 hours (self-reported; period and method not documented)

4

The source does not report the denominator of this figure.

Reported by
Coinbase
Scope
PR review times for the wallet rewrite effort
Observation date
2026-03
Key observation

Speed run: about 100 participants opened about 70 PRs in 15 minutes (opened, not merged; single event)

4 Supports5 Supports

The source does not report the denominator of this figure.

Reported by
Coinbase
Scope
Pull requests opened during one speed-run session
Observation date
2026-03

Lessons and interpretation

Support a portfolio of harnesses (Claude Code, OpenCode, Cursor, Copilot) rather than standardizing on one

1

Keep the system of record (Linear) as the agent's structured context; conversation can be the input, Linear stays durable

2

Simple per-agent worktree/branch/terminal isolation is a pragmatic alternative to full sandboxing for concurrent coding

1

Security requirements that blocked cloud background agents motivated the in-house, Slack-native harness build

5

Observation date
2026-03

Give background agents the same tools and context as engineers through a Slack-native interface

3

Observation date
2026

Spread adoption through visible team experiments, such as PR speed runs and shared Slack channels, rather than mandates

4 Supports5 Supports

Observation date
2026

Sources and research details

Citations link to the original publisher. Each source also keeps a preserved copy in the repository, so a changed or removed page stays checkable.

  1. Coding had a concurrency problem: how Mux helped solve ithttps://www.coinbase.com/de/blog/coding-had-a-concurrency-problem-how-mux-helped-solve-itEngineering blog · First party · Last source verification: 2026-08-31
    Preserved copy ↗
  2. Coinbase × Linear (Forge workflow)https://linear.app/customers/coinbaseCase study · First party · Last source verification: 2026-08-31
    Preserved copy ↗
  3. Chintan Turakhia: 5% of all PRs merged now come from our in-house background agentshttps://x.com/chintanturakhia/status/2019921786834411728Social post · Direct participant · Last source verification: 2026-09-15
    Preserved copy ↗
  4. How Coinbase scaled AI to 1,000+ engineers (How I AI, with Chintan Turakhia)https://www.lennysnewsletter.com/p/how-coinbase-scaled-ai-to-1000-engineersPodcast · Direct participant · Last source verification: 2026-09-15
    Preserved copy ↗
  5. How I AI: Chintan Turakhia's playbook for AI adoption at Coinbasehttps://www.chatprd.ai/how-i-ai/playbook-for-ai-engineering-adoption-at-coinbasePodcast · Direct participant · Last source verification: 2026-09-15
    Preserved copy ↗
Research details for every claim on this page
  1. Summary
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  2. Headline claim
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    Self-reported share stated by the engineering leader who built the harness; measurement method and period not documented. A thread reply says the share covers the whole company.
    Reported by
    Coinbase
    Scope
    PRs merged by in-house background agents
    Denominator
    All PRs merged company-wide
    Method
    Not reported
    Observation date
    2026-02
  3. Sandbox
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  4. Harness
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  5. Model
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  6. Interfaces
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  7. Tool access
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  8. Knowledge
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  9. Supporting component
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  10. Supporting component
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
  11. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    A linked first-party source states the claim.
    Reported by
    Coinbase
    Scope
    Registered Mux users including engineers, PMs, and designers; 335 active and 197 power users
    Denominator
    Not reported
    Method
    Not reported
    Observation date
    Not reported
  12. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    Self-reported share stated by the engineering leader who built the harness; measurement method and period not documented. A thread reply says the share covers the whole company.
    Reported by
    Coinbase
    Scope
    PRs merged by in-house background agents
    Denominator
    All PRs merged company-wide
    Method
    Not reported
    Observation date
    2026-02
  13. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Low
    Confidence reason
    The episode description states the reduction without a date, denominator, or measurement method.
    Reported by
    Coinbase
    Scope
    PR review times for the wallet rewrite effort
    Denominator
    Not reported
    Method
    Not reported
    Observation date
    2026-03
  14. Key observation
    Statement type
    Metric
    Provenance
    Reported
    Confidence
    Medium
    Confidence reason
    Described by the organizer as a single event; explicitly pull requests opened, not merged.
    Reported by
    Coinbase
    Scope
    Pull requests opened during one speed-run session
    Denominator
    Not reported
    Method
    Not reported
    Observation date
    2026-03
  15. Lesson
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    Medium
    Confidence reason
    The catalog derives this observation from the linked sources.
  16. Lesson
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    Medium
    Confidence reason
    The catalog derives this observation from the linked sources.
  17. Lesson
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    Medium
    Confidence reason
    The catalog derives this observation from the linked sources.
  18. Lesson
    Statement type
    Fact
    Provenance
    Reported
    Confidence
    High
    Confidence reason
    The episode writeup states the security requirement as the reason for the in-house build.
    Observation date
    2026-03
  19. Lesson
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    Medium
    Confidence reason
    The catalog derives this design guidance from the builder's description of the harness.
    Observation date
    2026
  20. Lesson
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    Medium
    Confidence reason
    The catalog derives this adoption guidance from the episode accounts of speed runs and shared channels.
    Observation date
    2026
  21. Operating model assessment
    Statement type
    Inference
    Provenance
    Catalog judgment
    Confidence
    High
    Confidence reason
    The source describes delegated implementation that returns a pull request and build for human review.
    Observation date
    2026

The catalog records no related implementation for this entry yet.

Back to the directory →