ProgRouter is an online routing framework for multi-agent LLM workflows. Instead of making one upfront query-level decision like cascade routing methods, it selects which LLM agent to call at each workflow step, adapting to evolving task progress, remaining difficulty, and cost or time budget constraints. It combines a multi-view progress scorer, a dual-path progress predictor, and an adaptive meta-gating mechanism to estimate the progress gain of each candidate model before deciding.
Stepwise LLM routing cuts multi-agent workflow costs while preserving quality
ProgRouter selects an LLM at each workflow step using estimated task progress, time constraints, and long-term cost efficiency.
Academic
Songyuan Li · Ahmed M. Abdelmoniem · Shiqiang Wang
Aston University · Queen Mary University of London · University of Exeter
Research Digest··2 min read
Thread:Multi-Agent Coordination
In one line
ProgRouter selects LLM agents at each workflow step using predicted task progress, cutting operating cost while keeping task-solving quality.
Before you read this
Three ideas this paper builds onmodel routingDirecting each query or subtask to the most suitable model to optimize cost, accuracy, or quality.75 papers →multi-agent coordinationThe collaborative operation of multiple AI agents to solve tasks more effectively or with greater capability.77 papers →runtime monitoringAudits an AI agent's live actions, internal states, or effects against constraints or policies to catch errors and attacks.107 papers →
Read it as
Li, Abdelmoniem, and Wang introduce ProgRouter, an online routing framework that adapts model selection as a multi-agent task unfolds rather than making a single decision at the outset.
The authors developed a multi-view progress scorer that combines broad workflow outcome categories with finer signals including subtask completion, progress trends, and the quality of the current workflow state.
The Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.