Harness Evolution as Learning: Approximation, Generalization, and Optimization Limits of Self-Improving Personal Agents
Zeyu Gan, Zixuan Gong and Yong Liu at the Gaoling School of AI, Renmin University of China, treat harness evolution for personal agents as a learning problem and analyze it with a preference benchmark plus approximation, generalization and optimization error bounds.
Ask this paper
Setting. The model is fixed; adaptation to a user's preferences happens only by changing the harness (context, memory, tools, execution).
Architecture. Explicit context handles some local preferences reliably, while computational mechanisms are needed for the more demanding ones.
Scale. Adding memory can make preference compliance go down as well as up in their construction.
Self-evolution. Several self-evolving recipes stay well below an oracle.
Theory. Each failure is mapped to an error term (reachable policies for approximation, capacity under finite interaction evidence for generalization, biased update dynamics for optimization), which gives a common account of what different harness-engineering practices try to reduce.
Abstract
As the capabilities of large language models (LLMs) continue to advance, increasing attention is turning to how to translate their abilities into useful behavior. Personal agents bring this question into everyday settings, where models are expected to serve individual users and continually adapt to their preferences. With the underlying model held fixed, such adaptation relies on harness engineering: designing and evolving the surrounding layer that manages context, memory, tools, and execution. Despite rapid progress, the factors governing effective harness evolution remain insufficiently understood. To narrow this gap, we investigate three central questions concerning harness architecture, harness scale, and self-evolution algorithms through complementary empirical and theoretical analyses. Empirically, we introduce a preference-oriented benchmark and systematically characterize the capabilities and limitations of personal agents associated with these three dimensions. Theoretically, we formulate harness evolution as a learning problem and explain these phenomena through approximation, generalization, and optimization errors. Analyses of reachable policies, capacity under finite interaction evidence, and biased update dynamics provide theoretical accounts of the observed phenomena. Together, these results offer a unified perspective on the limits of personalization through harness evolution and inform future harness design.