Forge, previously Claudebot, turns employee discussions into Linear issues, drafted fixes and pull requests returned for review. Coinbase separately uses Mux to host concurrent coding-agent workspaces.
Custom harness: Slack bug discussion → Linear issue → drafted fix → PR pushed back for review[Source 2, Source 5]
Where people stay involved
People review, approve, or guide the agent’s work here.
Reported: The documented checkpoint is the pull-request review: Forge pushes the PR back to Slack for review, and the orchestrating engineer reviews, gives feedback, and merges each parallel result.
Supervision level
Work-product review
A person reviews the draft or implementation.
3/5Level 3 of 5
Slack, GitHub, or Linear request → drafted fix and pull request pushed back for review[Source 1, Source 3]
The Slack agent Claude bot can use multiple underlying models despite its name; the documented portfolio spans harnesses (Claude Code, OpenCode, Cursor, Copilot), and no source names the models' providers[Source 1, Source 5]Multiple underlying models are stated; the documented portfolio spans harnesses, and no source names model providers.
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[Source 1, Source 2, Source 3, Source 5]
Sandbox
Not reportedThe sources describe separate git worktrees, branches, and terminals for concurrent agents in Mux, a separate platform; they do not establish this as the Forge execution environment.
Tool access
Linear treated as the structured product context / source of truth[Source 2, Source 5]
Knowledge
Linear as the durable structured-context layer for product work[Source 2]
Context management
Slack discussions become labelled, sized Linear issues that carry the work context; the feedback pipeline summarizes recordings into a reviewable issue before an agent acts[Source 2, Source 5]
Credentials
Not reportedNo capture documents credential handling for the agents; the security requirements that blocked cloud background agents are described without their mechanism.
Interfaces
Slack, GitHub, and Linear are named as invocation surfaces for Forge.
One person coordinates coding agents with separate worktrees, branches, and terminals[Source 1]
Orchestrate concurrent agents
Separate Mux context: An engineer runs several agents in parallel, each on its own worktree, and reviews, gives feedback, and merges as each finishes[Source 1]
Validation and failure handling
Not reported: No automated check of agent output is documented; the documented check is the human pull-request review cited with human involvement.
Reported observations
These results come from the sources. Notes below each result explain what it covers and any reported limits.
Headline claim
5% of all PRs merged come from in-house background agents (Feb 2026)[Source 3]
Reported by
Coinbase
Scope
PRs merged by in-house background agents
Denominator
All PRs merged; the scope beyond engineering is not documented in the preserved capture
Observed in February 2026
About this result
Category
Adoption output
Evidence type
Reported measurement
What it covers
Share of all merged pull requests produced by in-house background agents
Key observation
Mux: 600+ users including engineers, PMs, and designers (335 active, 197 power users)[Source 1]
Reported by
Coinbase
Scope
Registered Mux users including engineers, PMs, and designers; 335 active and 197 power users
The source does not report the denominator of this figure.
About this result
Category
Adoption output
Evidence type
Reported measurement
What it covers
Registered Mux users including engineers, PMs, and designers, with active and power-user counts
Key observation
Wallet rewrite PR review times reduced from 150 hours to 15 hours (self-reported; period and method not documented)[Source 4]
Reported by
Coinbase
Scope
PR review times for the wallet rewrite effort
The source does not report the denominator of this figure.
Observed in March 2026
About this result
Category
Cost latency
Evidence type
Reported measurement
What it covers
PR review times for the wallet rewrite effort
Key observation
Speed run: about 100 participants opened about 70 PRs in 15 minutes (opened, not merged; single event)[Source 4, Source 5]
Reported by
Coinbase
Scope
Pull requests opened during one speed-run session
The source does not report the denominator of this figure.
Observed in March 2026
About this result
Category
Adoption output
Evidence type
Reported measurement
What it covers
Pull requests opened during one speed-run session
Key observation
Mux: 5,068 merged PRs across 461 repositories and 10 orgs[Source 1]
Reported by
Coinbase
Scope
Mux merged pull requests across repositories and organizations
The source does not report the denominator of this figure.
About this result
Category
Adoption output
Evidence type
Reported measurement
What it covers
Mux merged pull requests across repositories and organizations
Key observation
Mux users show 3.5x more merged PRs per engineer than baseline (39.6 vs 11.4); the post cautions users may skew toward already-high-output engineers[Source 1]
Reported by
Coinbase
Scope
Merged pull requests per engineer for Mux users compared with a baseline of non-users
Denominator
Engineers not using Mux at the time of the comparison (11.4 merged PRs)
About this result
Category
Effectiveness
Evidence type
Reported measurement
What it covers
Merged pull requests per engineer for Mux users compared with non-user baseline
Lessons
Coinbase engineers use Mux alongside Claude Code, OpenCode, Cursor, and other internal agents.[Source 1]
Coinbase stores requirements and work status in Linear; its Slack workflow creates an issue before Forge drafts a fix.[Source 2]
Catalog interpretation: Mux gives concurrent agents separate worktrees, branches, and terminals. This separates edits; it does not establish a security sandbox.[Source 1]
Coinbase reports that security requirements prevented it from using available cloud background agents and motivated its internal harness.[Source 5]
Observed in March 2026
Coinbase describes its Slack harness as giving agents the same tools and context as engineers, with a PR returned to the thread.[Source 3]
Observed in 2026
Reported opinion: Chintan Turakhia favors demonstrations and shared experiments for adoption, citing PR speed runs and a Cursor Wins channel.[Source 4, Source 5]
Reported: The parallel-session Mux example remains in research details and has its own infrastructure record.
human involvement
Reported: The documented checkpoint is the pull-request review: Forge pushes the PR back to Slack for review, and the orchestrating engineer reviews, gives feedback, and merges each parallel result.
implementation
Reported
validation
Not reported: No automated check of agent output is documented; the documented check is the human pull-request review cited with human involvement.
observations
Not reported: The collected adoption metrics describe Mux, broader background agents or company experiments; they do not isolate Forge outcomes.
Self-reported share stated by the engineering leader who built the harness; measurement method and period not documented. The preserved tweet says only 'all PRs merged'; no preserved reply establishes a company-wide scope.
Reported by
Coinbase
Scope
PRs merged by in-house background agents
Denominator
All PRs merged; the scope beyond engineering is not documented in the preserved capture
5% of all PRs merged come from in-house background agents (Feb 2026)
Duplicate representation of Headline claim. Identical statement, period, scope, and denominator as the headline: the same 5% share of merged PRs from the same tweet.
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. The preserved tweet says only 'all PRs merged'; no preserved reply establishes a company-wide scope.
Reported by
Coinbase
Scope
PRs merged by in-house background agents
Denominator
All PRs merged; the scope beyond engineering is not documented in the preserved capture
Self-reported comparison whose own text cautions that Mux users likely skew toward engineers who were already high-output and AI-forward; no cohort method is documented.
Reported by
Coinbase
Scope
Merged pull requests per engineer for Mux users compared with a baseline of non-users
Denominator
Engineers not using Mux at the time of the comparison (11.4 merged PRs)
The source describes git and terminal separation but no access-control boundary between agents, so sandbox equivalence would be an unsupported inference.
