I gave two Claude Code sessions a PM and an engineer job. My job was to watch

Description and use

Two autonomous Claude sessions—PM and engineer—iterate through spec writing, implementation, review, and verification to produce a fully tested UI library with 128 closed tasks.

Problem

Automating a software development workflow with two AI agents—one acting as a product manager (spec writer and reviewer) and the other as an engineer (implementer)—to reduce human effort while ensuring code quality.

Outcome

The dual‑agent workflow produced a fully tested, continuously updated UI library with 128 closed tasks (115 bugs, 5 refactors, 3 performance issues, 5 notes), all verified by automated tests.

Workflow

• 1. PM session writes a component spec (requirements, scenarios, tasks).
• 2. Engineer session receives spec, implements code, and lists touched files in a message.
• 3. PM reviews implementation, writes findings as Reado comments anchored to code lines.
• 4. Engineer fixes each finding and marks it done with a task done --verify command.
• 5. PM verifies fixes; Reado records task as closed only if verification passes.
• 6. Repeat steps 3‑5 for up to two passes per change, then merge the change.

Evidence

Since August, Claude Code sessions can talk to each other. The feature is called cross‑session messaging and arrived in v2.1.224: a session finds your other sessions with ListAgents and sends one of them a message with SendMessage.

Source

The PM session writes each component as a spec (an OpenSpec change: the requirements, the scenarios, the tasks) and validates it before any code exists. When the engineer is done, the PM reviews the result.

Source

The engineer session takes one spec, implements it, and when it thinks it's finished, sends a message back asking for review. That hand‑off lists every file it touched, so the review knows exactly what to look at.

Source

The PM session doesn't write its findings into a chat message. It leaves them as Reado comments, anchored to the exact lines they're about, each one typed (bug, refactor, performance…) and marked as a task.

Source

So far the PM has left 128 of these tasks: 115 bugs, 5 refactors, 3 performance issues and 5 notes. All 128 are closed, and every one of them was closed by a verify command that passed.

Source

Two sessions reviewing each other's work produce good results, but they're expensive: every spec gets written, implemented, reviewed, fixed and verified, and each step reads a lot of code.

Source