Specialize Roles, Mix Deployments: Pushing the Cost-Accuracy Frontier of LLM Agent Teams

Summary

Treats cost and accuracy in multi-role LLM teams (planner, executor, verifier) as properties of the assignment of models to roles rather than of any single model, and introduces AgentCARD, a benchmark combining role-decomposed evaluation, unified cost modelling and Pareto-frontier analysis. Heterogeneous teams achieve up to 44% accuracy improvements or up to 12x cost reduction, and role bottlenecks vary by domain. The cleanest available treatment of the cost side of organizational design, which the paper needs if it claims a structure is better on epistemic quality but more expensive.

Contribution

Queued — not yet read.

Method

Queued — not yet read.

Relevance to RISE

Surfaced by the 2026-08-29 literature sweep as a medium-priority candidate for the Architecture as Epistemology (ISR) paper (angle: MAS failure and cost analyses; MAS topology / structure vs. aggregation rule). Verified against: Opened https://arxiv.org/abs/2606.20629 (arXiv abstract page, v1 28 May 2026)

Critique / open questions

Queued — not yet assessed.

Key quotes

LLM agents are increasingly deployed as multi-role teams, where tasks are divided across specialized roles such as planner, executor, and verifier. (abstract)

In these systems, cost and accuracy are no longer properties of a single model: they depend on which model fills each role and where it is hosted, including API, self-hosted, and hybrid deployment. (abstract)