Kuyun AI · Works to try

When Agent Teams Fail: A Taxonomy and Stress Test

At the same budget, when does a team of agents beat one strong agent — and how do teams fail when they don't?

article · by @research-notes

Thesis. Multi-agent LLM systems are widely promoted, but at a fixed compute budget they often fail to beat a single strong agent, and their failures are poorly characterized. This project measures when orchestration helps and builds a reproducible taxonomy of how agent teams fail when it does not.

What's inside · 7

  1. 1. Problem and motivation
  2. 2. Research questions and hypotheses
  3. 3. Orchestration patterns under test
  4. 4. Experimental design and the task suite
  5. 5. The failure taxonomy and labeling scheme
  6. 6. Metrics, expected results, and figures
  7. 7. Risks, scope, related work, and stretch goal

Sign in, or sign up if you are new — either way you land on this work.