Autonomous Internet Task Decomposition: Transforming Complex Human Intentions into Executable Multi-Capability Workflows

The increasing capabilities of generative artificial intelligence (GenAI), autonomous agents, search engines, and Internet-based services are creating new opportunities for systems that can accomplish complex objectives through natural-language instructions. However, many human requests cannot be completed through a single computational operation or an individual Internet capability. Instead, they require multiple interdependent activities, such as information retrieval, data collection, comparative analysis, content generation, external verification, and result synthesis. Despite advances in autonomous agents and large language model (LLM)-based planning, transforming an unrestricted human intention into a reliable, dynamically executable workflow across heterogeneous Internet capabilities remains an important research problem. This paper proposes Autonomous Internet Task Decomposition (AITD), a conceptual framework designed to transform complex human intentions into structured, executable, and dynamically adaptable multi-capability workflows. Rather than treating task decomposition as the simple conversion of a natural-language request into a list of subtasks, AITD conceptualizes decomposition as an intent-grounded process that identifies required outcomes, establishes hierarchical task structures, discovers execution dependencies, associates subtasks with suitable Internet capabilities, and constructs an adaptive execution graph. The central contribution of AITD is the introduction of a capability-oriented dynamic task execution graph that connects human intentions, hierarchical subtasks, heterogeneous Internet capabilities, execution dependencies, intermediate outputs, and final outcome requirements within a unified representation. The framework incorporates intent-to-task transformation, hierarchical task representation, capability-aware task assignment, dependency-aware execution, parallel and sequential scheduling, dynamic task expansion, failure detection and recovery, subtask completion verification, and overall goal satisfaction assessment. Unlike approaches that primarily emphasize action selection, conversational reasoning, or predefined agent workflows, AITD places particular emphasis on the structural relationship between complex human objectives and the heterogeneous Internet capabilities required to accomplish them. It further distinguishes between a task being attempted, a task being executed, a task producing an output, and a task being successfully verified as contributing to the intended outcome. This paper develops the conceptual architecture, formal representation, operational mechanisms, potential advantages, research propositions, evaluation methodology, and implementation considerations of AITD. It also examines the relationship between AITD, hierarchical task-network planning, autonomous agents, workflow orchestration, and intent-driven human–Internet interaction. Rather than claiming that task decomposition or autonomous planning has not previously been investigated, this work seeks to establish a capability-oriented Internet execution perspective in which complex human intentions become the organizing basis for dynamically generated and verifiable multi-capability workflows.