Workspace Optimization: How to Train Your Agent

TL;DR

This paper introduces workspace optimization for agents with fixed language models, focusing on evolving external structured workspaces to improve multi-turn task performance.

Contribution

It proposes a novel method for evolving an agent's external workspace, mirroring weight-space training, and demonstrates its effectiveness with the DreamTeam multi-agent system.

Findings

DreamTeam improves ARC-AGI-3 score from 36% to 38.4%.

DreamTeam uses 31% fewer environment actions per game.

The approach is effective in hard multi-turn environments.

Abstract

Modern agents built on frontier language models often cannot adapt their weights. What, then, remains trainable? We argue it is the agent's \emph{workspace}, the structured external substrate it reads, writes, and tests; we call its evolution workspace optimization. Workspace optimization targets hard multi-turn environments where a frontier model has strong priors but cannot solve the task in a single shot, so the agent must learn through interaction. We propose a principled way to evolve the workspace, mirroring the structure of weight-space training: artifacts in place of parameters, evidence in place of data, counterexamples in place of losses, and textual feedback in place of gradients. We instantiate the idea in DreamTeam, a multi-agent harness for ARC-AGI-3 whose roles build an executable world model, plan, hypothesize, probe, strategize, and route failures. On the current 25-game ARC-AGI-3 public set under the official scoring protocol and averaged over two independent runs, DreamTeam improves the SOTA protocol-matched agent's score from 36% to 38.4%, while using 31% fewer environment actions per game.