AutoAgent: A Fully-Automated and Zero-Code Framework for LLM Agents
Jiabin Tang, Tianyu Fan, Chao Huang
TL;DR
AutoAgent is a fully-automated, zero-code framework that enables anyone to create and deploy LLM agents using natural language, significantly lowering the technical barrier for AI agent development.
Contribution
It introduces AutoAgent, a novel autonomous agent operating system that allows non-technical users to build and customize LLM agents through natural language commands.
Findings
Outperforms state-of-the-art methods on the GAIA benchmark.
Demonstrates superior retrieval-augmented generation capabilities.
Enables dynamic tool and workflow creation without manual coding.
Abstract
Large Language Model (LLM) Agents have demonstrated remarkable capabilities in task automation and intelligent decision-making, driving the widespread adoption of agent development frameworks such as LangChain and AutoGen. However, these frameworks predominantly serve developers with extensive technical expertise - a significant limitation considering that only 0.03 % of the global population possesses the necessary programming skills. This stark accessibility gap raises a fundamental question: Can we enable everyone, regardless of technical background, to build their own LLM agents using natural language alone? To address this challenge, we introduce AutoAgent-a Fully-Automated and highly Self-Developing framework that enables users to create and deploy LLM agents through Natural Language Alone. Operating as an autonomous Agent Operating System, AutoAgent comprises four key components:…
Peer Reviews
Decision·ICLR 2026 Conference Withdrawn Submission
- The paper reframes agent development as zero-code, language-first engineering, combining native tool-use with XML-style calls and event-driven workflows in a novel, practical way. - It delivers a coherent end-to-end “agent OS” with executable rigor (tool auto-testing, coding sandbox) and strong results on GAIA and MultiHop-RAG. - The architecture is cleanly modularized with explicit interfaces and traceable, step-by-step creation logs that make reproduction straightforward. - By removing codin
- The claim of being the “first” or uniquely natural-language–driven is not well substantiated, since many agent frameworks are already prompt- or language-driven. - The paper reads more like a carefully engineered stack than a conceptually new algorithm. - The chosen name is easily confusable with existing systems claiming automatic agent generation[1], while this work still relies on a library of predefined agents/tools and a fixed orchestration backbone. - Despite zero-code claims, the approa
1. The paper is well motivated. It aims to address an important accessibility issue in LLM agent development by removing the coding barrier for non-technical users. 2. The proposed framework integrates multiple functional modules into a coherent and extensible agentic operating system. 3. The method achieves competitive results on GAIA and state-of-the-art RAG scores, demonstrating its effectiveness.
The paper's motivation is to make agent creation accessible to laypeople. However, this is not reflected in the evaluation. The experiments are all on standard benchmarks, without any realistic or interactive evaluation. Also, there is no user-centered experiment or study showing the usability for non-technical users. Besides, in my opinion, the paper sometimes overstates its contributions, e.g., calling it “revolutionary” when much of the technical content builds upon established frameworks.
1. AutoAgent addresses a critical gap in LLM agent frameworks by eliminating the need for coding skills, making agent technology accessible to a broader audience. The natural language-driven approach—from agent creation to tool integration—represents a significant step toward inclusive AI development. 2. The framework's decomposition into four core components enables flexibility and scalability. The Agentic System Utilities facilitate seamless collaboration between specialized agents (e.g., Web,
1. The framework relies heavily on external LLMs (e.g., GPT-4, Claude) for core reasoning, which may introduce costs, latency, and reliability issues. The paper does not explore the impact of LLM quality variations or fallback mechanisms for open-source models, potentially affecting consistency in production use. Missing reference: Junyu Luo et al., Large language model agent: A survey on methodology, applications and challenges, arXiv 2025.
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsMulti-Agent Systems and Negotiation · Digital Rights Management and Security · Access Control and Trust