NetGent: Agent-Based Automation of Network Application Workflows

NetGent is a research paper (arXiv:2509.00625v2) presenting an agent-based system for automating network traffic data collection. Key defensibility constraints: (1) Zero adoption signals (0 stars, 0 forks, 219 days old, no release velocity) indicate this is pre-release academic work with no public repository traction. (2) The core innovation—using LLM agents to automate browser workflows for realistic traffic generation—is a novel_combination of existing components (LLMs, browser automation, network packet capture) but lacks technical depth that creates switching costs. (3) Platform domination risk is HIGH: This directly competes with emerging capabilities in major cloud providers (AWS, GCP, Azure) who are building native synthetic traffic generation, chaos engineering, and ML dataset tooling. OpenAI/Anthropic could trivially add agentic workflow automation to their platforms. (4) Market consolidation risk is MEDIUM: Incumbents like Perfecto, BrowserStack, and cloud-native load testing platforms (k6, JMeter, Locust) could absorb this as a feature within their own agent-orchestration layers. (5) The displacement horizon is 1-2 years because: platforms are actively investing in agentic AI (OpenAI agents, Anthropic Claude Tools); browser automation-as-a-service vendors are commodity now; and network traffic synthesis is a well-understood problem. The only defensibility would come from a large, domain-specific dataset of labeled traffic or a robust open-source community—neither exists yet. (6) Integration surface is reference_implementation: this is academic code accompanying a published paper, likely suitable for reproduction but not hardened for production use. (7) Composability as a component is theoretical—the system could be embedded in larger ML pipelines, but without a pip package, API, or stable release, integration friction is high. The project lacks the community, network effects, or technical moat needed to survive well-funded platform expansion.