50 AI agents had 3 hours to build. They formed a committee.

WHAT HAPPENED

In Minecraft, a 50-agent team never thrived

Researchers asked 50 agents to build a thriving civilization in three hours. After the first night, many ambitious tasks were abandoned. Four agents spent 11 minutes appointing a "sole fertilizer operator" for one piece of bone meal.

THE NUMBER

11 min

4 agents picking one operator for one piece of bone meal

Source: Tom's Hardware report on the Minecraft swarm test.

THE REAL PROBLEM IN COMPANIES

What I see in companies adding agents to a workflow

1. 01Every new agent adds hand-offs, and nobody owns the end result
2. 02Roles overlap, so agents debate who decides instead of doing the work
3. 03No one measures output per agent, so the overhead stays invisible

I'VE ALREADY SOLVED THIS

Multi-agent in production at an enterprise account

1. 01A multi-agent platform running in production at a US$5.5 million account
2. 02About 60% efficiency gain from the platform
3. 03Multi-agent in production requires an audited defect log

MY TAKE

Simple that works beats complex that impresses.

The Minecraft test is a game, but in my reading the failure mode is familiar. Agents without tight roles and limits produce process, not output. In production I start small, define who does what, and measure what comes out.