Work

A Tri-Mode Agent-Aware Internet Architecture with Verifiable Human–Agent Semantic Consistency: A Unified Framework for Human-Web, Agent-Optimized, and Dual-Representation Access

Year: 2026
Type: preprint
Abstract: The rapid emergence of autonomous and semi-autonomous artificial intelligence (AI) agents is challenging the long-standing assumption that Internet resources are primarily accessed and interpreted by human users. Conventional web infrastructures predominantly deliver information through human-oriented interfaces, including HTML pages, visual layouts, images, videos, and interactive elements. Although such representations remain essential for human communication, they can impose unnecessary processing costs and structural ambiguity on AI agents that retrieve, interpret, compare, and reason over online information. Recent developments in agent-oriented web design have introduced machine-readable representations, agent-specific interfaces, and content negotiation mechanisms. However, the coexistence of human-oriented and agent-oriented representations raises a fundamental architectural question: how can an Internet resource support different access preferences without allowing its representations to diverge in factual content, contextual meaning, or temporal state? This paper proposes a Tri-Mode Agent-Aware Internet Architecture (TMAIA) with a central emphasis on verifiable human–agent semantic consistency. The architecture defines three explicitly selectable access modes: Human-Web Only, Agent-Optimized Only, and Dual-Representation Access. The third mode enables an agent to retrieve both the conventional human-readable representation and its corresponding machine-optimized representation within a coordinated access operation. Unlike a simple format-selection mechanism, the proposed framework treats representation correspondence, shared resource identity, version synchronization, provenance, and consistency verification as architectural requirements. It introduces an adaptive representation router, a common authoritative resource model, a dual-delivery coordinator, a semantic consistency engine, and a versioned provenance mechanism. The framework also distinguishes agent identification from representation selection: an agent should be able to declare its preferred access mode explicitly rather than depending exclusively on server-side detection. This paper presents the conceptual architecture, operational model, protocol-level design considerations, security requirements, and an evaluation framework. It does not claim experimental validation. The proposed contribution is a unified architectural formulation of three-mode access and verifiable cross-representation consistency, building upon rather than replacing existing Internet standards and agent-oriented web proposals.
Language: English
Open Access status: green
OpenAlex ID: W7225919607
Cites:
Cited by:
Related to: 10