PM and engineer agents in one room
Product and designPM in Claude Code and engineers in Codex, same repo
The PM decided the export limit was 50,000 rows. The engineer's Codex built for one million. When the engineer found that streaming exports time out at the proxy after 30 seconds, the PM's spec still promised 'any size'.
How do I make sure my engineers' Codex follows the product decisions I make in Claude Code?
Put both agents in the same ArcRoom project room. Your decisions reach the engineer's Codex through recall, and the engineer's failed approaches, saved with remember_failure, reach your next Claude Code session. The team dashboard shows both agents' sessions on the project. A live view of what each agent is doing right now is on the roadmap; today it's what they saved, plus the dashboard.
Also asked as: The engineer built for a million rows when the spec said 50,000. How do we keep agents in sync? · How do PMs and engineers share context between Claude Code and Codex? · Can my Claude Code session see what failed in an engineer's Codex?
Both agents use the same project room.
- Failed approach
- Decision
- Saved decision
- Recall
- Dashboard
Recall (engineer, Codex): Decision (pm-lena): export limit 50k rows in v1 (why: support load)
Step by step
- Both agents use the same project room.
- The PM's decisions reach the engineer's Codex through recall.
- The engineer's failed approach is saved with remember_failure and reaches the PM's next session.
- The team dashboard shows both agents' sessions on the project.
What the agent sees
Recall (engineer, Codex): Decision (pm-lena): export limit 50k rows in v1 (why: support load)
Recall (PM, Claude Code): failed: streaming export (proxy timeout at 30 s, codex-arun, Oct 30)Compared with today's tools
Slack thread + shared spec doc
PM and engineer post decisions in Slack and paste them into their agents.
- Setup
- None
- Ongoing
- Paste and chase, ~10 min a day
ArcRoom
Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.
- Setup
- 2 min per agent (npx arcroom init --agent claude / codex)
- Ongoing
- None
Mem0's OpenMemory could share one store between the two agents, but only facts they choose to save; ArcRoom also passes the engineer's failed approach and its reason to the PM's next session through recall, without either agent searching.
Where it stops
Live visibility of what each agent is doing right now is on the roadmap. Today it's what they saved, plus the dashboard.
Try it in your project
Works with the agents your team already uses, and any agent over MCP or the HTTP API.