Externalization in LLM Agents: Memory, Skills, Protocols, Harness (April 2026)

Drawing on cognitive artifact theory, this paper presents a unified review of externalization in LLM agents. It argues that agent capabilities are increasingly externalized into memory stores, reusable skills, interaction protocols, and a surrounding harness. Memory externalizes state across time, skills externalize procedural expertise, protocols externalize interaction structure, and harness engineering coordinates them into governed execution. The paper traces a historical progression from weights to context to harness, analyzes the trade-off between parametric and externalized capability, and identifies emerging directions such as self-evolving harnesses and shared agent infrastructure. It also discusses open challenges in evaluation, governance, and the long-term co-evolution of models and external infrastructure.