Use case · A mixed-agent engineering team

One team.
Three agents.
One way to ship.

How a team leader brings Codex, Claude Code, and Cursor users into one governed workflow — while every developer keeps control of their own agent, account, and machine.

Follow the journey ↓

↔AGENT PAIR
SHARED ROOM

TLTeam leadergoverns the flow

Codeximplements

Claude Codereviews

Cursorships a change

THE SCENARIO01

A team can standardize the workflow without standardizing the tool.

Meet Maya, a team leader working on a checkout service. Alex prefers Codex, Sam works in Claude Code, and Jordan uses Cursor. Agent Pair gives them one shared project, one source of truth, and clear ownership from task definition to pull request.

The complete journey

From a leader's intent
to a trusted change.

Every handoff is explicit. Every result is tied to the task that produced it. Every external action waits for the right approval.

SET UP

Maya creates the room

She connects the repository, defines project policies, and invites the team. The project becomes a shared Agent Room with durable membership, tasks, artifacts, and decisions.

Checkout reliability3 members · 3 agents · GitHub connected

CONNECT

Everyone brings their own agent

Alex, Sam, and Jordan install a local connector and authenticate their preferred provider on their own machines. Credentials stay local; Agent Pair only sees the declared capability and connection status.

CodexAlex

Claude CodeSam

CursorJordan

ASSIGN

A bounded task goes to the right agent

Maya writes the outcome and acceptance criteria, then assigns implementation to Jordan's Cursor agent. The task has a clear owner, action level, and approval requirement.

● ASSIGNEDHarden payment retriesOwner · Jordan / Cursor · Tests required

→

EXECUTE

The agent works in an isolated worktree

Cursor receives only the bounded task context. Agent Pair creates the worktree, runs the connector, tracks progress, and captures a commit, patch, tests, assumptions, and risks with provenance.

Room Run in progresscursor-agent · payment-service · 68%

12:48

REVIEW

A second agent checks the exact result

Maya assigns review to Sam's Claude Code agent. Claude receives the produced artifact version, runs its checks, and returns a review report rather than silently changing the work.

Claude Code review12 files · 34 tests passed

READY

APPROVE & SHIP

Maya approves what she actually reviewed

Once the result and review look right, Maya approves that immutable artifact version. Agent Pair can now prepare the GitHub pull request — and refuses if the approved version changes.

✓

Human approval recordedartifact v3 · ready for pull request

Different tools, shared contract

What each provider does.
What Agent Pair owns.

The providers stay distinct at the edge. The collaboration model stays consistent at the center.

Local providerBest suited forAgent Pair guarantees

CodexImplementation, refactoring, test-driven changesBounded task context, isolated worktree, provenance

Claude CodeReview, reasoning over a produced artifact, risk findingReview is attached to the exact artifact version

CursorFast implementation in a familiar editor workflowLocal credentials, governed execution, safe replay

The trust boundary

Local control.
Shared accountability.

Agent Pair coordinates the work without taking ownership of anyone's provider account. The team gets a reliable delivery process; each developer keeps their preferred environment.

ON EACH MACHINEProvider accountLocal connectorGit worktreeCredentials never cross this boundary

⇄signed
interop

IN THE SHARED ROOMTasks & ownershipArtifacts & reviewsApprovals & auditOne canonical workflow

Ready for a team workflow?

Give every agent
a place in the process.

Explore the source ↗Back to Agent Pair ↗

One team.Three agents.One way to ship.