Skip to slide
Chapter 3 · The Case Against: Cognition's "Don't Build Multi-Agents"
19 / 41

CHAPTER 03 · The Case Against: Cognition's "Don't Build Multi-Agents" · 3 / 9

The fragile architecture, illustrated with Flappy Bird

Yan takes aim at a tempting and common design:

  1. The agent breaks the work into parts.
  2. It starts subagents to work on those parts in parallel.
  3. It combines the results at the end.

He calls this fragile, and uses a concrete example. Suppose the task is "build a Flappy Bird clone." It gets split into Subtask 1, "build a moving game background with green pipes and hit boxes," and Subtask 2, "build a bird you can move up and down."

Things go wrong. Subagent 1 misreads its subtask and builds a background that looks like Super Mario Bros. Subagent 2 builds a bird, but it does not look like a game asset and it moves nothing like Flappy Bird. Now the final agent is stuck trying to combine two pieces that do not fit. The example may sound contrived, but Yan's point is that real tasks carry many layers of nuance, and each layer is a chance for a subagent to misunderstand.

← → arrow keys work too