Ecdysis speeds up AI-agent harness training, a paper summary reports

A post summarizing Ecdysis says it repairs recurring failure patterns across tasks, with several diagnostic roles agreeing on a change before any code is modified.

1 Source, 39d ago, first seen 39d ago

TLDR

A post sharing the Ecdysis paper describes a method for improving agent harnesses—the supporting code around AI agents. It highlights two problems: testing changes is slow, and fixes can overfit by treating model failures as harness bugs. According to the summary, Ecdysis analyzes failures across tasks and repairs only recurring patterns, after several diagnostic roles agree on the change. The post reports training up to 1.84x faster than existing harness evolution methods and an 18.56% gain in reasoning accuracy for the resulting harnesses. It also reports better transfer across large language models, fewer tokens used and results matching full-data training with a quarter of the data.

Today's Rank

—

Not ranked yet

Combined views

—

1 Source, first seen 39d ago

1 Source

@dair_aiGreat paper on self-evolving agent harnesses. Self-evolving agent harnesses have two practical problems: 1. Search is slow, because every candidate harness needs repeated agent runs and code edits. 2. Fixes overfit, because each failure is patched as if it were a harness bug, even when the model caused it. Ecdysis analyzes failures across a batch of tasks and repairs only patterns that recur. Several diagnostic roles agree on a change specification before any code is modified. Harness training runs up to 1.84x faster than existing harness evolution methods, and the resulting harnesses gain 18.56% in reasoning accuracy. They also transfer better across LLMs, use fewer tokens, and match full-data training with a quarter of the data. Paper: https://arxiv.org/abs/2609.11677 Chat with Paper: https://academy.dair.ai/papers/ecdysis-efficient-and-effective-training-of-runtime-harnesses-for-llm-agents-2609.1167739d

Today's Rank

—

Not ranked yet

Combined views

—

1 Source, first seen 39d ago

— likes

— comments

— saves

— reposts

— comments

— saves

— reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Today's Rank

—

Not ranked yet

Today's Rank

—

Not ranked yet

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

1 Source

@dair_aiGreat paper on self-evolving agent harnesses. Self-evolving agent harnesses have two practical problems: 1. Search is slow, because every candidate harness needs repeated agent runs and code edits. 2. Fixes overfit, because each failure is patched as if it were a harness bug, even when the model caused it. Ecdysis analyzes failures across a batch of tasks and repairs only patterns that recur. Several diagnostic roles agree on a change specification before any code is modified. Harness training runs up to 1.84x faster than existing harness evolution methods, and the resulting harnesses gain 18.56% in reasoning accuracy. They also transfer better across LLMs, use fewer tokens, and match full-data training with a quarter of the data. Paper: https://arxiv.org/abs/2609.11677 Chat with Paper: https://academy.dair.ai/papers/ecdysis-efficient-and-effective-training-of-runtime-harnesses-for-llm-agents-2609.1167739d