Agentic Web: A Research Repository Mapping AI Agent Studies for the Next-Generation Web
Agentic Web: Weaving the Next Web with AI Agents.
At a glance
- What is it?
- The Agentic Web repository from SAIL Research Lab is a curated collection of papers and resources on AI agents operating in web environments, accompanying the survey paper published at arxiv.org/pdf/2507.21206. It covers agentic web development, information retrieval, recommendation, agent planning, multi-agent learning, safety, and benchmarks.
- Who is it for?
- The Agentic Web repository is the right starting point for researchers who want a structured overview of the academic literature on AI agents interacting with web environments. It does not contain runnable code or benchmarks directly; researchers who need evaluation tooling must follow the citations in the README to the specific benchmark repositories listed there.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 76 days ago.
- What is it written in?
- GitHub does not report a main language for this repository.
Answers come from the project's GitHub data, last synced on October 1, 2026, and from our analysis. They are not legal advice.
Editorial analysis
What the Agentic Web Is and Why This Repository Exists
The agentic web refers to the emerging paradigm in which AI agents, rather than human users, perform tasks on the web: searching for information, filling forms, navigating interfaces, making recommendations, and coordinating with other agents. This is a departure from the Web 2.0 model, where humans drove all interactions and the web was built for direct consumption.
The README frames this through a three-era timeline: Web 1.0 (directories and static pages), Web 2.0 (user-generated content and social platforms), and the emerging agentic web where AI agents act as participants rather than tools. It notes that these eras are not strictly separated and that technologies overlap during transitions.
The repository is associated with a survey paper whose preprint is at arxiv.org/pdf/2507.21206. The authors are affiliated with Shanghai Jiao Tong University, the Hong Kong University of Science and Technology Guangzhou, the University of Liverpool, UC Berkeley, Shanghai Innovation Institute, UC Davis, Virginia Tech, and University College London. The corresponding authors are Shangding Gu (UC Berkeley), Weinan Zhang (Shanghai Jiao Tong University), and Jun Wang (University College London).
How the Repository Is Organized
The repository top level contains a LICENSE file, README.md, and a docs/ directory. The content is in the README, which is organized into seven thematic sections:
- Agentic Web Development: papers on the broader infrastructure and architecture of the agentic web, including blockchain-enabled trust systems and AI agent registries - Information Retrieval: papers on how agents search and retrieve information from web sources - Recommendation: papers on agent-driven recommendation systems - Agent Planning: papers on how agents decompose tasks and plan sequences of actions - Multi-Agent Learning: papers on coordination and communication between multiple agents - Safety and Security: papers on adversarial inputs, trust, and safe agent behavior - Benchmark: papers presenting evaluation suites for agentic web tasks
Each entry in the README lists the paper title with a link, the authors, and the year. The list spans papers from 2022 through 2025.
Coverage: Key Papers and the Scope of the Literature
The paper list covers several sub-communities within agentic web research. In agent planning, the README includes WebArena (2023), which provides a realistic web environment for building autonomous agents; WebDancer (2025) on autonomous information seeking; and the ReAct framework (2023) on synergizing reasoning and acting. SWE-bench (2023) on resolving real GitHub issues is listed under Agentic Web Development.
In multi-agent learning, Toolformer (NeurIPS 2023) and ToolLLM (covering 16,000 real-world APIs) appear alongside more recent papers on multi-agent retrieval-augmented generation and chain-of-thought reasoning.
The safety section covers threat modeling and defense strategies for agentic systems. The benchmark section points to evaluation datasets that researchers can use to measure agent performance on web tasks.
The README includes a citation entry for the survey paper itself, which researchers who use this resource in their work are asked to cite.
Agentic Web Search and Browsing in the Paper Coverage
A substantial portion of the repository addresses how agents perform web search and browsing autonomously, which the README terms agentic web search. Papers in the information retrieval section cover retrieval-augmented generation, deep research agents, and web search agents that combine search with reasoning.
The Deep Research Agents paper (2025) listed in the README examines agents that perform multi-step research tasks by iteratively searching, reading, and synthesizing web content. MA-RAG (2025) covers multi-agent retrieval-augmented generation using collaborative chain-of-thought reasoning.
From Web Search towards Agentic Deep Research (2025) is listed in the Agentic Web Development section and discusses incentivizing search agents with reasoning. These papers collectively describe the architecture of agentic web search: agents that do not just retrieve a document but plan a search strategy, follow links, and synthesize findings across multiple sources.
This cluster of papers is the most directly relevant to practitioners building AI search pipelines or research automation tools.
Limitations of a Living Repository
The README describes the repository as under active development and acknowledges that it will be incomplete. The paper list reflects what the authors have indexed at the time of each commit; it will miss papers published after the last update and may miss papers the authors are unaware of.
Contributions are accepted by pull request, issue, or email to [email protected]. The inclusion process has no documented peer review or quality filter; any paper the community considers relevant to agentic web research can be submitted. This means the list may vary in quality across entries.
There is no code, dataset, or evaluation framework in the repository itself. The docs/ directory is present but its contents are not described in the README. Researchers who need to run experiments must source all tooling from the papers cited.
The Apache-2.0 license permits reuse and redistribution of the repository's text content with attribution.
Comparison with AwesomeLLMAgents and Related Survey Lists
The AwesomeLLMAgents and similar awesome-list repositories on GitHub collect papers on LLM-based agents broadly, including coding agents, embodied agents, and tool-using agents. These lists tend to be maintained by the community rather than by a single research group and grow through pull requests from many contributors.
The Agentic Web repository is narrower in scope: it focuses specifically on agents interacting with web environments (browsing, search, recommendation, web APIs) rather than all categories of LLM agents. This makes it more useful for researchers in the web and information retrieval communities and less useful for researchers working on robotics or code generation agents.
The connection to a specific survey paper also provides a conceptual framework for the collection. Where awesome-list repositories are essentially flat catalogs, the Agentic Web README groups papers by functional role in the agentic web ecosystem, making it easier to find papers relevant to a specific component of the system a researcher is building.
Editorial conclusion
The Agentic Web repository is the right starting point for researchers who want a structured overview of the academic literature on AI agents interacting with web environments. It does not contain runnable code or benchmarks directly; researchers who need evaluation tooling must follow the citations in the README to the specific benchmark repositories listed there. The last push was on 2026-07-18, and the repository is under active development according to the README. Contributions are welcomed via pull request, issue, or email to the corresponding authors.
Frequently asked questions
What is the agentic web?
According to the repository, the agentic web is the emerging paradigm in which AI agents perform web tasks autonomously rather than humans doing so directly. It represents a shift from Web 2.0's human-driven interaction model toward agents that search, retrieve information, make recommendations, and interact with web interfaces on a user's behalf.
What is agentic web search?
Agentic web search refers to AI agents that plan and execute multi-step search strategies, follow links, and synthesize findings across multiple sources, rather than returning a single document. Papers in the repository covering deep research agents and retrieval-augmented generation address this pattern directly.
Where is the survey paper for Agentic Web?
The companion survey paper is available at arxiv.org/pdf/2507.21206. The repository README also provides a citation entry for researchers who use the resource in their own work.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/sail-research-lab-agentic-web)