Agent Organization (Experimental)

Ante supports multiple patterns for organizing agents to work together. Each architecture trades off between autonomy, coordination overhead, and result quality.

Independent

Agents work in parallel on the same problem with no interaction. An aggregator synthesizes their outputs at the end.

Best for: tasks where diverse independent perspectives improve quality (brainstorming, redundant verification).

Decentralized

Agents run in parallel rounds, reading each other's prior outputs and proposing refinements. After a fixed number of rounds, consensus is formed without a central coordinator.

Best for: debate-style reasoning, peer review, or negotiation where no single authority should dominate.

Centralized Iterative

A central orchestrator decomposes the problem, dispatches agents in parallel, evaluates their results, and decides whether to refine or finish.

Best for: complex tasks that benefit from top-down planning with quality gates (code generation with review, multi-step research).

Hybrid Iterative

Combines centralized orchestration with decentralized peer refinement. The orchestrator plans and dispatches agents, then agents refine each other's work in a peer round before the orchestrator evaluates.

Best for: high-quality collaborative output where both structured planning and peer feedback matter (collaborative writing, architecture design).

Choosing an architecture

Architecture	Coordination	Iteration	Use when
Independent	None	Single pass	You need diverse perspectives without interaction overhead
Decentralized	Peer-to-peer	Fixed rounds	Agents should self-organize without a central authority
Centralized Iterative	Orchestrator-driven	Quality-gated	You need structured decomposition with evaluation checkpoints
Hybrid Iterative	Orchestrator + peers	Quality-gated	You want both top-down planning and bottom-up peer refinement