From 2026 to 2030, advances in AI and wireless communications technology are expected to bring many changes to several core areas of engineering practice.

Agentic AI and standardised protocols are expected to simplify engineering workflows. A hybrid structure integrating non-terrestrial networks (NTN) and terrestrial networks (TN) will expand wireless communications coverage. New AI-based technologies will improve embedded systems and simulation processes. These trends are expected to fundamentally change how engineers design, connect and manage complex systems.

Engineering workflows redefined by agentic AI and the Model Context Protocol (MCP)

In engineering, the next stage of AI evolution is agentic AI. Unlike existing large language models (LLMs) that respond only based on internal knowledge, agentic AI systems can gather additional information or run tools that automate tasks. These systems can select the appropriate tools for a user request, format data to suit the tool and post-process results. This enables agentic AI systems to create and modify files, run code and resolve errors, opening up vast possibilities.

Developers are now strengthening capabilities to safely integrate agent-based AI into practical work. Today’s agentic AI systems are most effective when using a limited number of tools, but research is under way to expand their ability to select and use more tools. As these systems are granted access rights to file systems, databases and code execution to perform software development tasks, ensuring security becomes more important. LLMs can make errors, but ongoing research is focused on reducing security risks and making agentic AI’s powerful capabilities safer and more reliable. If these efforts bear fruit, they will lay the groundwork for broader adoption of agentic AI and a large impact across industries.

For AI agents to function properly, they need a reliable way to accurately understand and exchange information. The Model Context Protocol (MCP) addresses this by standardising how tools, data and prompts are shared among the components of agentic AI systems. By standardising communication and context, MCP reduces the scope for misunderstanding and enables smooth collaboration across tools and teams. Engineers can choose among various MCP-integrated tools to build the best combination of tools suited to each problem’s characteristics.

When AI agents and MCP are combined, engineering models can be interpreted and manipulated regardless of the software used or organisational boundaries. This makes it possible to propose design alternatives, coordinate simulations and adjust engineering workflows in real time to align with project goals and industry standards. As these technologies mature, engineers will be able to spend less time on tool and data management and invest more time in creative problem-solving.

Hybrid NTN-TN networks to gain full momentum in 2026

As non-terrestrial networks enter a full-scale deployment phase, implementation cases are emerging that complement terrestrial network infrastructure. The 3GPP Release 17 standard sets reliability and latency parameters and presents a benchmark for NTN-TN interoperability. Release 18 provides an essential foundation for implementing scalable high-throughput architectures through NTN-IoT and support for high-frequency bands. For wireless engineers, these changes are bringing new design and integration challenges across direct-to-cell connectivity and overall network coordination.

NTN does not replace TN. It reinforces it to form a hybrid ecosystem that will define next-generation global wireless connectivity. A key technical challenge for wireless engineers is ensuring stable switching between satellite and terrestrial links. Because handover management and resource coordination ultimately determine the success or failure of overall system design, interoperability between NTN and TN is more important than anything else. For the RF industry, the emergence of NTN-TN networks means rising demand for flexible multiband transceivers and robust channel modelling that can operate across diverse propagation environments.

AI driving performance improvements in complex embedded systems

AI’s influence on embedded software is accelerating. Complex embedded systems have traditionally relied on rule-based logic and manual tuning algorithms. A shift is now under way toward deploying advanced AI models directly to microcontrollers, FPGAs, GPUs and NPUs. This integration will enable edge devices to make faster and smarter decisions locally, reducing reliance on cloud connectivity and improving system resilience.

Three key technologies support the shift to embedded AI models: model compression, automated code generation and system-level model testing. Structural model compression methods such as pruning and projection, along with data type compression methods such as quantisation, enable complex models to run efficiently on edge devices. Automated code generation tools convert compressed AI models into C and C++ code optimised for platform-specific implementations. System-level model testing ensures compressed and deployed models operate reliably within the overall embedded system and verifies functional accuracy and real-world behaviour. These tools help engineers move quickly and confidently from concept to deployment.

One application area where embedded AI models play a central role is virtual sensing. Engineers use AI models to infer and estimate physical quantities that are difficult or costly to measure directly, based on other sensor data. This approach increases monitoring efficiency and reduces the need for additional hardware sensors while maintaining accuracy and reliability. Mercedes-Benz has introduced embedded AI technology to develop a deep learning-based virtual sensor for real-time mass flow estimation and installed it directly in an ECU. Advances in virtual sensing are enabling more intelligent and responsive systems across a broad range of embedded applications while lowering costs and complexity.

Innovation in simulation and design processes through AI-based ROM

As the scale and complexity of engineering challenges increase, AI-based reduced order modelling (ROM) technology is expected to become more common. AI-based ROM is helping engineers bridge the gap between detailed first-principles simulations and the need for fast design exploration, optimisation and real-time simulation. AI-based ROM enables models to be computationally efficient while maintaining strong predictive power.

AI-based ROM simplifies complex physics-based models by capturing the most essential dynamics using computationally efficient neural networks or other AI architectures. This allows engineers to run simulations and optimisation faster, making it possible to analyse complex systems in real time. One example is a pure black-box AI model trained only on input and output data from a high-fidelity simulation model. There are also hybrid physics-informed machine learning models that leverage engineers’ physics knowledge. Because hybrid models can be trained with less data, they can reduce the number of expensive full-order model simulation runs. Hybrid models also excel at generalisation, providing more reliable predictions across various input signals and parameter values. AI-based ROM will change many industries, including automotive, aerospace and energy. In automotive engineering, ROM can help optimise charging strategies, extend battery life and strengthen safety functions in electric and hybrid vehicles. AI-based models capture electrochemical dynamics to improve battery management systems (BMS) and allow control systems engineers to run fast system-level simulations to verify BMS logic. AI-based ROM can also help aerospace engineers predict aerodynamic forces and structural responses during flight. It reduces computational requirements to enable real-time simulation, supports lighter and more efficient aircraft designs, and accelerates material performance tests without extensive wind tunnel experiments. The energy sector uses ROM to predict equipment performance and system behaviour that are essential for grid stability and predictive maintenance. ROM also predicts failures of critical components such as transformers and turbines.

Wireless channel modelling evolving with generative AI

Wireless engineers are now exploring ways to use LLMs within workflows and design. Researchers are considering how to use LLMs to enable context-aware decision-making and simplify the management of complex wireless environments.

One important process that could benefit from LLM integration is wireless channel modelling. Originally seen as an auxiliary function with limited scalability, accurate channel modelling has become an essential process for multi-user multiple-input multiple-output (MIMO) and beamforming systems. Generative AI will support engineers in generating more realistic and practical channel models by enabling them to explore complex scenarios that were previously difficult to implement.

It is still too early for LLMs to directly control physical layer functions such as beam steering, but they can provide information to guide higher-level decisions that direct RF operations. In the early deployment stage there will be constraints on power and computing performance, but ongoing research on lightweight generative AI models and AI-native architectures will make scalable implementations possible that can run on edge devices. For wireless system designers, these changes indicate a growing need to integrate physical layer performance with AI-based orchestration and decision-making.