From the source
Agent sessions work well for focused tasks, but most real projects are too broad and complex for a single context window to hold.
A single agent eventually runs into a problem: the more it sees, the less focused and reliable it becomes.
Missions is our system for solving that.
It breaks large work into focused units handled by fresh agents with narrowly scoped goals, shared state, and explicit validation.
Rationale Most of the architecture follows from one core observation: agents are highly reactive to their context.
An agent's trajectory is append-only, so the model's reasoning at any given point is a function of every past thought, observation, and action.
Models seek coherence: they integrate what's in their context into a unified worldview and reason forward from it.
Therefore, they perform best when every previous step in the trajectory urges them toward the next optimal step.
When the context window accumulates information that is irrelevant to - or actively working against - the current goal, performance suffers.
Two failure modes follow from this: Irrelevant context accumulates.
…
