# Paper warns self‑improving agents may fail when skill sets change

> Researchers introduce CODA, a test that shows co‑evolved AI agents lose most of their gains when their skill library drifts after deployment.

Oossa · 2026-10-16 · https://oossa.com/en/paper-warns-self-improving-agents-may-fail-when-skill-sets-change

A team led by Jinluan Yang posted a new arXiv paper on Oct 16, 2026. It shows that agents trained to improve both their policy and a skill bank become fragile if the skill bank changes later. The authors propose a new evaluation called CO‑evolution Drift Audit (CODA) that simulates skill drift and measures consistency and resistance. Across four methods and five benchmarks, CODA finds worst‑case success drops 5‑13 points, wiping out most of the advantage the agents had over a no‑skill baseline.

## The facts

- Paper posted Oct 16, 2026
- Worst‑case success 5‑13 points lower than peak, erasing 52‑93% of gain

## Why it matters

Deployers will need to test AI agents for robustness to skill changes, not just peak performance.

## Sources & references

1. [Evolving Together, Failing Apart: The Skill-Policy Co-evolution Trap in Self-Improving Agents](https://arxiv.org/abs/2610.20845) – arXiv, 2026-10-16

Last updated: 2026-10-16