cd /news/ai-agents/two-people-two-coding-agents-one-rep… · home › topics › ai-agents › article
[ARTICLE · art-170112] src=dev.to ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

Two people, two coding agents, one repo: what goes wrong, and what I built for it

A developer built agentroom, a shared coordination room for teams running multiple coding agents (Claude Code, Codex, Cursor, Antigravity) on the same repository. The tool lets agents claim files before editing, share goals, specs and decisions, maintain a searchable team memory of dead ends, distribute approved skills, and cross-review each other's finished tasks, connected via hooks and a single MCP server. The developer notes it does not merge code or catch logically conflicting changes that merge cleanly, and file claims are per file rather than per line.

by read2 min views1 publishedNov 15, 2026

If you and a teammate each run a coding agent (Claude Code, Codex, Cursor) on the same repo, the agents have no idea the other one exists. I kept running into the same four problems: #

Same file, two agents. Both edit it. You find out at the merge. #

Duplicate work. Both pick up the same task, or one rebuilds a helper the other finished an hour ago. #

No shared memory. My agent learns that a test setup is a dead end. My teammate's agent learns it again the next day. #

Rules that live on one laptop. A skill or checklist I wrote for my agent never reaches theirs.

Git worktrees fix the first one on a single machine. They don't help across two laptops, and they do nothing for the other three.

#

What I built

agentroom is a shared room for a team's agents. Each person connects their agent with a room code, and from then on:

Agents claim files before editing. If another agent holds a file, the edit is blocked and the agent is told who has it and for which task. It can pick other work, wait, or raise a conflict. #

Goals, specs and decisions are shared. Every agent starts its session knowing what the team decided. #

There is a team memory. Agents write down dead ends and where things live, and search it before exploring. It is searched on demand, so it doesn't sit in every session's context. #

Skills can be shared. One person shares a skill with the room, the host reads every file and approves it, and the other agents get it installed. #

Agents review each other. With review turned on, a finished task waits for another agent's verdict, ideally from a different model.

It works across Claude Code, Codex, Cursor and Antigravity through hooks and one MCP server, so nobody has to switch tools.

#

What it doesn't do

It doesn't merge code for you, and it doesn't catch two changes that merge cleanly but disagree in logic. File claims are per file, not per line. Review is advisory: it doesn't block a merge. I'd rather say that up front.

#

  Try it

It's in early access: [request a room](https://agentroom-app.vercel.app/access). If you get stuck, email me at [talukapunya.1@gmail.com](mailto:talukapunya.1@gmail.com).

I'd like to hear how your team handles this today, especially if you've found something simpler that works.

── more in #ai-agents 4 stories · sorted by recency
── more on @agentroom 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/two-people-two-codin…] indexed:0 read:2min 2026-11-15 · —