Repeated Labor Market Allocation

Agents bid on tasks using self-estimated private execution costs and self-maintained strategy notes; a central allocator selects winners from bids plus public reputation records and applies a VCG-style payment rule that rewards honest cost-aware bidding. Complex tasks are handled through hierarchical delegation: a winning agent can decompose its task and subcontract pieces back through the same bidding market rather than a planner assigning subtasks directly.

Properties

family: Coordination & Scaling
maturity: frontier
diagram: negotiation
whyMultiAgent: A market only exists among distinct bidders holding separately private cost estimates; a single agent has no competing cost estimate to bid against and no reputation record distinct from its own. The paper's own comparison shows a single inserted preference can nearly double a favored agent's task share under a centralized LLM allocator, a manipulation vulnerability specific to a scheme where one planner assigns work to many distinct agents, which is exactly what the decentralized bidding market is built to remove. Distinct from the already-tracked market-belief-trading pattern, where price signals aggregate independently held probabilistic beliefs toward a truth-seeking consensus: here the market allocates task work via cost bids and VCG payments, not beliefs, and there is no truth-seeking objective. Also distinct from negotiation-equilibrium, where two agents with opposing utilities negotiate directly over shared state: here many agents compete through an allocator via one-shot sealed bids plus reputation, with no direct agent-to-agent negotiation.
mast: [Cat 1]
useCases: [Orchestrating large heterogeneous pools of third-party LLM agents with unknown or self-reported costs][Decentralized task allocation resistant to a single planner's manipulable preferences][Hierarchical subcontracting of complex tasks across specialized agents]
source: Markets, Not Planners: Decentralized Orchestration of LLM Agents with Private Information · arXiv:2608.23867

Provenance

Source: https://arxiv.org/abs/2608.23867

Repeated Labor Market Allocation

Properties

family

Coordination & Scaling

maturity

frontier

diagram

negotiation

whyMultiAgent

A market only exists among distinct bidders holding separately private cost estimates; a single agent has no competing cost estimate to bid against and no reputation record distinct from its own. The paper's own comparison shows a single inserted preference can nearly double a favored agent's task share under a centralized LLM allocator, a manipulation vulnerability specific to a scheme where one planner assigns work to many distinct agents, which is exactly what the decentralized bidding market is built to remove. Distinct from the already-tracked market-belief-trading pattern, where price signals aggregate independently held probabilistic beliefs toward a truth-seeking consensus: here the market allocates task work via cost bids and VCG payments, not beliefs, and there is no truth-seeking objective. Also distinct from negotiation-equilibrium, where two agents with opposing utilities negotiate directly over shared state: here many agents compete through an allocator via one-shot sealed bids plus reputation, with no direct agent-to-agent negotiation.

mast

[Cat 1]

useCases

[Orchestrating large heterogeneous pools of third-party LLM agents with unknown or self-reported costs][Decentralized task allocation resistant to a single planner's manipulable preferences][Hierarchical subcontracting of complex tasks across specialized agents]

source

Markets, Not Planners: Decentralized Orchestration of LLM Agents with Private Information · arXiv:2608.23867