Exo: An Architecture for Self-Improving Systems at Runtime
cs.AI
Submitted: 2026-10-30
Updated: 2026-10-30
Code: https://github.com/MaxBittker/rs-sdk
License: http://creativecommons.org/licenses/by/4.0/
The gist: Agent harnesses govern how language models act in an environment; designing their policies by hand is difficult, while automated techniques require representative datasets and costly executions.
Abstract
Agent harnesses govern how language models act in an environment; designing their policies by hand is difficult, while automated techniques require representative datasets and costly executions. We present Exo, an agent runtime architecture for safe learning during ongoing work that collapses the optimization loop into task execution, allowing agents to revise both their harness and their strategies for further improvement. A minimal protected mechanism layer preserves execution history and governs policy activation while the harness evolves through inspectable code changes. On long-running objectives in an open-world game, Exo reduces model cost by 12-53% relative to Codex and Pi in standalone runs and 45-82% in transfer runs, while matching task completion rates. Compared with the same starting policy without the ability to edit policy, Exo reduces held-out model cost by 35-61% on AppWorld and SREGym with comparable correctness, cutting AppWorld's mean task time by 13% and SREGym's median diagnosis time by 35%. On Continual Learning Bench coding tasks, model cost including learning falls by 14%, with both policies resolving every issue.
Sources
- GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning
- Continual Learning Bench: Evaluating Frontier AI Systems in Real-World Stateful Environments
- LogAct: Enabling Agentic Reliability via Shared Logs
- Trace is the Next AutoDiff: Generative Optimization with Rich Feedback, Execution Traces, and LLMs
- SREGym: A Live Benchmark for AI SRE Agents with High-Fidelity Failure Scenarios
- Continual Harness: Online Adaptation for Self-Improving Foundation Agents
- Meta-Harness: End-to-End Optimization of Model Harnesses
- A Self-Defining Operator for Distributed Applications
- AutoHarness: improving LLM agents by automatically synthesizing a code harness
- TTHE: Test-Time Harness Evolution
- AlphaEvolve: A coding agent for scientific and algorithmic discovery
- ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory
- Measuring Agents in Production
- Reflexion: Language Agents with Verbal Reinforcement Learning
- Self-Healing Harness for Runtime Oversight of Agent Self-Modification
- Live-SWE-agent: Can Software Engineering Agents Self-Evolve on the Fly?
- RRSI: Regularized Recursive Self-Improvement of Agent Harnesses
- Harness 2: Recursive Agent Harnessing for an Open World
- ReAct: Synergizing Reasoning and Acting in Language Models
Related papers
- Multi-Objective Structured Pruning of LLMs for Latency and Model Size Optimization
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Embeddings for Preferences, Not Semantics
- Can We Trust Item Response Theory for AI Evaluation?
- Diagnosing Task Insensitivity in Language Agents