Alibaba’s SkillWeaver framework cuts AI token consumption by over 99%

Alibaba researchers have built a framework called SkillWeaver that slashes token consumption by more than 99% when AI agents tackle complex, multi-step workflows.

The core problem SkillWeaver addresses is deceptively simple: when an AI agent has access to hundreds or thousands of tools and skills, it struggles to pick the right one for each step.

How SkillWeaver actually works

Traditional tool-routing systems use what’s called a one-shot approach. The agent looks at a task, scans its entire library of available skills, and picks tools in a single pass.

SkillWeaver takes a fundamentally different approach. It constructs an execution graph for a given task using directed acyclic graphs (DAGs), essentially mapping out all the subtasks and their dependencies before choosing any tools. Then it applies a technique the researchers call Skill-Aware Decomposition, or SAD, which uses an iterative feedback loop to fetch, evaluate, and select the most relevant tool candidates for each node in the graph.

Under the hood, the framework pairs a dependency-aware DAG planner with an LLM decomposer and a bi-encoder retriever built on FAISS, the open-source similarity search library originally developed at Meta.

The foundational paper, titled “SkillWeaver: Web Agents can Self-Improve by Discovering and Honing Skills,” was submitted on April 9, 2025. It features co-authors from Ohio State University, University of Virginia, Purdue University, Carnegie Mellon University, and Cisco Research. The paper describes web agents that autonomously create reusable skills designed as APIs.

Why a 99% reduction in token use matters for markets

Tokens are the fundamental unit of compute cost for large language models. Every API call to GPT-4, Claude, or any other frontier model is priced per token. If an enterprise AI workflow previously consumed 100,000 tokens per task execution, SkillWeaver could theoretically bring that down to under 1,000.

Alibaba’s broader AI ambitions and the crypto angle

SkillWeaver doesn’t exist in isolation. Alibaba has been aggressively building out its AI infrastructure, including the Qwen family of open-source language models and its Wukong platform. The company’s AI Agent Skills Portal serves as a centralized hub for managing the kinds of tools and skills that SkillWeaver is designed to route more efficiently.

SkillWeaver itself has no direct connection to blockchain or cryptocurrency. The research documentation makes no mention of tokens in the crypto sense, decentralized infrastructure, or on-chain applications.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.