Intel / article

Subagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic Tasks

Source: arxiv.org
Author: Wasu Top Piriyakulkij, Rachel Lawrence, Alicia Curth, Sushrut Karmalkar, Niranjani Prasad
Date: Published Sep 10, 2026

Context

Reusable skills bundle instructions, scripts and supporting material for a task. This study compares loading those instructions into the main agent's conversation with assigning the to a that starts with its own fresh .

The authors report an advantage for subagents when a package clearly specifies its inputs and required outputs and contains the procedure needed to produce them. Those boundaries let the delegated agent concentrate on a defined subtask without carrying the main conversation's accumulated material. Coordination consumes additional tokens, so the authors frame it as a tradeoff, with the benefit depending on how a skill is organized and invoked.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.

Key quotes

“We show that subagent execution outperforms agent-skill execution when skill packages expose clear input-output contracts and their instructions encode the procedural knowledge needed to fulfill those contracts.”

“Our results show that the benefit of reusable knowledge depends not only on its content, but also on how it is organized and invoked.”

Read the source arxiv.org