Agentic Web

• Agentic Web: Weaving the Next Web with AI Agents

The emergence of AI agents powered by large language models (LLMs) marks a pivotal shift toward the Agentic Web, a new phase of the internet defined by autonomous, goal-driven interactions.

• The web is evolving from human-driven interaction to machine-to-machine interaction where agents plan, coordinate, and execute complex tasks autonomously
• The paper proposes a structured framework built around three key dimensions: intelligence, interaction, and economics

• Intelligence — the cognitive layer of individual agents
  • Retrieval: accessing and synthesising web-scale information
  • Recommendation: personalised, context-aware suggestions
  • Planning: decomposing goals into executable sub-tasks
  • Collaboration: coordinating with other agents or humans
• Interaction — how agents communicate and coordinate
  • Communication protocols between agents (see AI Communication Protocols)
  • Orchestration strategies for multi-agent pipelines (see Agentic Design Patterns)
  • Tool and API integration with existing web infrastructure
• Economics — incentive structures governing agent behaviour
  • Agent Attention Economy: agents competing for computational resources, API bandwidth, and user attention in a market-like fashion
  • Economic incentives for cooperation vs. competition among agents
  • Micropayment and value-exchange primitives for autonomous transactions (see Payment Protocols)

• Communication protocols — standardising how agents expose capabilities and exchange messages (see AI Communication Protocols)
• Orchestration — managing task delegation, context passing, and failure recovery across agent networks
• Scalability — maintaining coherent behaviour as agent networks grow
• Integration — fitting autonomous agents into existing HTTP/web service infrastructure without breaking backward compatibility

• Concept introduced by the paper to describe resource allocation in multi-agent environments
• Agents must compete or negotiate for: API rate limits, LLM inference slots, human oversight bandwidth
• Creates emergent market dynamics analogous to advertising auctions on today’s web
• Raises questions about fairness, monopolisation, and optimal mechanism design

• Web automation — end-to-end task execution across browser-accessible services
• Information retrieval & synthesis — multi-hop research across live web sources
• Scientific research assistance — hypothesis generation, literature review, experiment design
• Data analysis pipelines — agents chaining retrieval, computation, and reporting
• Autonomous commerce — price negotiation, procurement, fulfilment

• Paper
  • Agentic Web: Weaving the Next Web with AI Agents
    • 17 authors incl. Pieter Abbeel, Dawn Song, Weinan Zhang, Jun Wang
    • 2025-07-28
    • arXiv: cs.AI, cs.LG