A 48-hour hackathon team

Agencies and freelancersThree people, three agents, one weekend

At 3 a.m. two teammates' agents both rebuilt the login screen.

How do we stop our AI agents from building the same thing twice at a hackathon?

Everyone runs arcroom init in the repo at the start, then agents claim parts with ArcRoom's claim_task so work doesn't overlap. If a part is taken, the agent sees who has it, like 'already claimed by dev-2 (cursor)'. For demo prep, saved answers show what each part does. Setup costs a few minutes at the start.

Also asked as: Two of our agents rebuilt the login screen at 3 a.m. How do we prevent that? · How do I split work between teammates' AI agents on a short project? · Can teammates' AI agents see which part someone else already claimed?

The work

Everyone runs arcroom init in the repo at the start.

ArcRoom keeps
  • Claim
The next agent gets

claim_task → already claimed by dev-2 (cursor): "login screen"

Step by step

  1. Everyone runs arcroom init in the repo at the start.
  2. Agents claim parts, so work doesn't overlap.
  3. The final demo prep finds what each part does from saved answers.

What the agent sees

claim_task → already claimed by dev-2 (cursor): "login screen"

Compared with today's tools

Shared task board (Trello or GitHub issues)

The team lists the parts on a board and each person moves a card before telling their agent to start.

Setup
15 min
Ongoing
remember to move cards at 3 a.m., ~1 min per task

ArcRoom

Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.

Setup
2–5 min per person
Ongoing
None

A board works when people update it; claim_task checks inside the agent, so a duplicate is caught even when someone forgets.

Where it stops

Setup costs a few minutes at the start.

Try it in your project

Works with the agents your team already uses, and any agent over MCP or the HTTP API.

More in agencies and freelancers