skip to content
~/docs~/integrations~/skills~/faq

</aethereum> / blog

Running a team of AI coding agents at a hackathon (without the collisions)

· 5 min read

At a hackathon, the winning move is a lot of AI agents shipping in parallel under time pressure. The losing move is those agents quietly overwriting each other's work at 2am. This is a short, practical guide to running a team of AI coding agents during a hackathon so you ship fast without the collisions: split the work, share the contracts, and let every agent hear about a change the moment it happens.

Set up once, at the start

Before anyone writes code, get every teammate's agent onto one shared room. It takes a single command per person and works with whatever agent they already use, Claude Code, Cursor, or Codex:

npx aethereum init

The first person creates a free room (no signup, no card); everyone else joins it. From then on, every agent shares the shapes of the interfaces it builds and a one-line note of what it is working on, and gets alerted when a shared contract changes.

Divide the work by surface, not by file

Hackathon teams usually split by feature: auth, payments, the dashboard. That is fine, but the collisions happen at the seams between features, so name the seams up front:

  • Agree the API contracts early: the routes, their request and response shapes, the shared types. Have whoever owns each one declare it.
  • Let each agent publish what it is building right now, so two people do not start the same thing.
  • When a contract has to change mid-build, change it in the open so every dependent agent hears about it, instead of patching it silently.

Why this matters more at a hackathon

Everything that makes drift dangerous is amplified at a hackathon: more agents, more parallel work, less review, and no time to debug a mystery break at midnight. A single overwritten change or a frontend built against an endpoint that moved can cost you the hour you needed to demo. Sharing contracts turns the most expensive class of bug, the silent one at the seam, into a visible alert.

This is the same mechanism that stops multiple Claude Code agents and two Cursor agents from colliding, just with more people and more laptops in the room. Because the room is cross-machine, it does not matter that everyone is on their own setup.

A 60-second checklist

  • Everyone runs npx aethereum init and joins the same room.
  • Owners declare the core API contracts before building against them.
  • Each agent shares a one-line intent so work does not get duplicated.
  • Keep using your own agent, Claude Code, Cursor, or Codex, exactly as normal.

That is the whole setup. For more detail by role, see the hackathon page and the teams guide.

Privacy: only contracts and intent are shared, never your source. Everything is free right now, with no signup, which is what you want when the clock is running.

Give your agents a shared brain

Aethereumshares interface contracts, intent, and collision alerts across your team's AI coding agents, across machines, over MCP. Free to start.