In this space last month, I wrote about the hot buzzwords at June’s IMS in Boston, one of which was “artificial intelligence.” Now, I don’t know about you, but outside of the context of my professional life, I’ve grown weary of hearing about AI.

The pop-culture perception of AI represents something less than honorable. Social media is drowning in “AI slop.” The hallucinations of large language models (LLMs) often end up in legal filings, sometimes to judges’ horror and/or amusement. And in case you missed it, an autonomous AI agent created by OpenAI recently went rogue and hacked a tech startup.

And yet, the building of enormous AI data centers continues apace, as do the burgeoning valuations of AI-centric businesses. If real monetization of AI isn’t achieved relatively soon, it’s easy to envision a burst tech bubble not unlike the dotcom one of some 25 years ago.

Not All AI Results in Slop

But when I do look at AI in the context of my professional life, I feel considerably better about it. In our world of wireless systems design, AI presents a much more likable face and brims with potential.

I recently moderated a webinar titled “Future Directions for AI and ML in RF/Microwave Design Tools.” In presentations by Alexander Petr of Keysight EDA and Giorgia Zucchelli of MathWorks, we learned where AI is today in design flows. We also gained some fascinating insights into how it can impact RF designers’ productivity in the future while maximizing the impact of the analog design intuition that good engineers bring to their jobs.

Both Keysight EDA and MathWorks have augmented their RF/microwave design flows with agentic AI to transform a largely manual design flow into one that greatly accelerates automated exploration of the design space.

Each of these analog EDA giants brings their own spin to the subject. Keysight maps a journey toward RF design autonomy from manual GUIs to what it terms “organizational intelligence” that leverages institutional domain knowledge and continuous learning.

Keysight’s approach relies on Python scripts that become reusable functions or building blocks. Once a workflow is captured, it's then able to be automated at scale. A parallelized “design swarm” can be thrown at a problem, rapidly running through the multitude of alternatives within the design space to find the optimal solution.

For its part, MathWorks describes two paths in using AI for RF/microwave design, both of which center on MATLAB. In one, custom solutions can be used to solve specific problems and execute well-defined tasks, such as antenna design. For example, AI can start with very sparse data, such as a pair of orthogonal slices (azimuth and elevation) of an antenna’s far-field radiation pattern.

From that 2D representation, MATLAB can reconstruct the 3D pattern, predicting its performance through a neural network. When the results are compared to those obtained with electromagnetic simulation, they’re surprisingly close.

Bringing Agentic AI to Bear on Complex Tasks

The other path described in MathWorks’ presentation is a fuller use of agentic workflows to investigate open-ended problems and execute highly complex tasks. In the webinar (which is archived on our site), you can watch as an RF signal chain is built using the designer’s specifications.

The AI agent can direct MATLAB to run a budget analysis using code it builds. The agent creates a Simulink model of the RF chain and applies a testbench. In natural language, the engineer guides the AI agent through iterations and refinements using “what-if” scenarios.

Both vendors’ AI-assisted flows rely on similar technologies. In each case, the agentic-AI workflows are enabled by Model Context Protocol (MCP) technology that enables any tool to be connected to any AI agent. Both flows also rely on “skills,” which take the form of markdown files that tell the agent how and when to perform tasks, so they’re done correctly.

In my estimation, MathWorks seems to be a bit ahead of the game in enabling natural-language input to an AI agent, but both are working along similar lines.

What’s exciting about this new generation of agentic-AI-assisted design tools is that they enormously accelerate the design process. This is accomplished largely through use of surrogate models, which approximate EM models but still provide a similar result in a fraction of the time of a fixed EM simulation. You can test early and often, leading to quicker convergence on a solution.

Moreover, the designer remains at the center of the process. It’s your specifications, your reasoning with the AI agents, and your design expertise and intuition that drive the process. The AI agents simply command the tools to execute your intentions. Meanwhile, these futuristic flows provide a path toward leveraging AI technology without an engineer needing to be an AI expert.

So, let’s all ignore the online AI slop and focus on the exciting things that are happening in the latest generations of RF/microwave design tools. It’s a genuinely productive use of AI technology that won’t make you want to throw your phone away.