ActiveSaddler: Automated Curriculum Learning for Agent Harness Optimization
A new method called ActiveSaddler automates the optimization of LLM agent harnesses by adapting the training curriculum to the evolving needs of the harness.
01
Who it touches
- 1ActiveSaddler
- enables →Fact
- depends on →Fact
- uses →Fact
02
Evidence
- AarXiv cs.AIPrimary source2026-10-02 12:00
We formulate this missing dimension of harness optimization as an automated curriculum learning problem and introduce ActiveSaddler.
- AarXiv cs.AIPrimary source2026-10-02 12:00
ActiveSaddler models the evolving curriculum as a non-stationary bandit with dynamically instantiated optimization targets.
- AarXiv cs.AIPrimary source2026-10-02 12:00
However, existing methods primarily optimize how the harness is updated while largely fixing which training scenarios generate the feedback that drives those updates. As the harness evolves, the scenarios most useful for further optimization can change, suggesting that the training curriculum itself should adapt alongside the harness. We formulate this missing dimension of harness optimization as…