.jpg)
Edge computing and artificial intelligence converge at a critical point: resource management in 5G and 6G networks for vehicle-to-everything (V2X) communications. Current packet scheduling systems must guarantee latencies on the order of milliseconds for services such as teleoperated driving or sensor sharing. Here, large language models show limitations in speed and reliability, paving the way for hybrid architectures where small language agents act as non-real-time policy generators, while lightweight controllers execute validated decisions. This approach, called Agentic-V2X, proposes a balance between semantic flexibility and operational determinism.
The proposal relies on a locally deployed small language model that produces policies for priorities, weights, and safety constraints based on a scenario summary, service objectives, and telemetry. A validator corrects and verifies these policies before an xApp-like controller applies them through dynamic adaptation of scheduler weights. This design overcomes the latency and hallucination limitations of large LLMs, but does not aim to be the best at all operating points; experimental results over 126 runs in ns-3 show that the adaptive solution competes favorably in critical reliability under high density, although it does not outperform the most robust static policies in aggregate. This positions it as a safe and executable architecture, not a universally dominant scheduler.
For a company like Q2BSTUDIO, this paradigm illustrates how artificial intelligence for businesses can be integrated into mission-critical systems without losing control. Our expertise in custom applications and custom software allows us to build modular platforms where AI agents collaborate with deterministic decision engines. Additionally, we offer AWS and Azure cloud services to deploy these agents in edge environments, and cybersecurity to protect V2X communications. We also develop business intelligence services with Power BI to monitor performance KPIs, and automate complex processes through specialized AI agents.
The main lesson from Agentic-V2X is that artificial intelligence in next-generation networks does not need to be omnipotent; it only needs to be useful, verifiable, and safe. Combining small language models with logical validators is a strategy that transcends the telecommunications domain and applies to any field where contextual adaptation is required without sacrificing guarantees. At Q2BSTUDIO, we apply this philosophy when developing AI solutions for businesses that act as decision assistants, not black boxes.
The study also underscores the importance of systematic experimentation: 126 runs with metrics such as deadline-constrained reception ratio, queue latency, deadline violations, throughput, fairness, and policy validity. This rigor is what we demand in our custom application projects, where each component is validated before integration. If your organization seeks to implement hybrid AI architectures for dynamic environments, we invite you to explore how our capabilities in custom software development can bring these concepts to life. For example, in the context of V2X networks, policy management can benefit from custom software that adapts agents to each operator's specific requirements, ensuring the solution is not only technically viable but also aligned with business objectives.
