Principle 2: Actions carry implicit decisions

Even after sharing the original task, the Flappy Bird system can still fail. This time you might get a bird and a background that work individually but clash visually, because Subagent 1 and Subagent 2 could not see what the other was doing. Each made reasonable assumptions, but the assumptions conflicted. This leads to the second principle:

Principle 2: Actions carry implicit decisions, and conflicting decisions carry bad results.

The insight is that every action an agent takes silently encodes a decision (a visual style, a data format, an interpretation). When two agents act in parallel without seeing each other's choices, their hidden decisions collide, and the collision shows up only at the end, when it is expensive to fix.

Yan argues these two principles are so important, and so rarely worth violating, that you should by default rule out any architecture that breaks them. The parallel-subagent pattern breaks them, which is why he distrusts it.