March 4, 2026
story
Now, distinguished jurors (in the embedded industry), I’m just a poor country lawyer (editor) and I don’t buy into the novelty LLM stuff that the bigwigs (you know who they are) are so passionate about. But the way I see it, as we move into the embedded world, it’s time to realize that we’re doing some serious work in the embedded industry, and maybe we should put away the toys.
Leaving aside the extremely thin frame device, let me make my passionate opinion clear. LLMs are pointless, harmful, unsustainable toys for marketing teams and not worth an engineer’s time. These consume resources that could be used for real-world solutions such as machine learning or processing AI. LLM-AI in general and Generative AI in particular have repeatedly caused failures, delays, legal risks, security and privacy risks, and large unrecoverable costs to nearly every system deployed. [Editor’s Note: This column originally appeared in the Spring 2026 issue of Embedded Computing Design Magazine.]
Here are some well-known, not-so-secret examples of these problems.
ChatGPT and OpenAI are losing money on every query, and most financial experts agree that the companies are surviving thanks to injections of outside money, while producing (if I’m being generous) factually questionable results on their queries. And now ads are being added.
Grok is used to create legally questionable adult content and monetize GenAI’s image capabilities.
Google claims it has no plans to include advertising in Gemini, but a recent Adweek report says investors are being told a different story. Remember, previously search was also not colored by ads.
As I warned in 2023, code written by GenAI is very likely to incur patents, intellectual property, and legal liability that could undermine real-world value: https://embeddedcomputing.com/technology/ai-machine-learning/using-generative-ai-for-code-can-be-a-big-risk
Environmental impact is a big issue, and a PR crisis is completely avoidable.
Companies really need to consider whether the marketing value of jumping on the GenAI and LLM chatbot hype is worth the risk. Perhaps this explains the 95% launch failure rate for AI pilots that MIT reported last year. Or maybe your LLM just isn’t working.
emphasize positivity
Let’s leave the LLM where it belongs – behind us – and consider more positive AI tools.
It’s unsettling to hear that companies are looking to leverage LLMs from the edge. This is because there is currently no use case for it. Last year was all about edge AI, and we saw some amazing innovations in powerful, compact processing and energy efficiency. This innovation gives me hope. That’s exactly what I mean when I talk about “practical AI tools.”
Small language models can be very useful at the edge and elsewhere in the work chain when trained on specific datasets for specific uses. We encourage engineers to focus on these SLMs and how they work to power operations at the edge, in vehicles, in factories and warehouses, and even in homes and hospitals. (Privacy and security need to be taken seriously, but that’s a topic for another column.)
The hottest thing in the AI field right now (and the last thing I’ll mention in this screed) is physical AI. NVIDIA’s Jensen Huang spent most of his keynote at CES talking about CES, but now every analyst and enterprise marketer wants to talk about how they’re leveraging physical AI.
Hey. Physical AI is just embedded computing.
We can all agree on that, right? No new terminology was needed. You’ve been creating this “physical AI” throughout your career. Huang is trying to carry on what Cisco abandoned in 2013 when it unsuccessfully rebranded IoT as the “Internet of Everything.” And just because NVIDIA is the (apparent) leader in AI chips, I don’t think it should be allowed to function.
Big AI has no real future. Consumers are moving away from that. Companies are abandoning it. If we believe that AI is truly valuable to business operations, makes the world easier to navigate, and (though that idea dies) actually improves the living conditions of people around the world, we should forget about the hype cycle. Products solve problems, not create them.
It’s time to put away the LLM and GenAI toys. Not included in actual product. As we continue to innovate and develop SLM and smart embedded engineering from the edge to the server, from the factory to the refrigerator, we have no doubt that the pilot-to-product ratio will move in a different direction.
I will rest my case.
Ken Briodagh is a writer and editor with 20 years of experience. He loves technology and if he had a teacher he would beta test everything from shoe phones to flying cars. In previous lives, he was a rush order cook, a telemarketer, a medical supplies technician, a funeral home body mover, a pirate, a poet, an alliterator, a parent, a partner, and a usurper to various thrones. Most of his accomplishments are exaggerated or outright false.
Ken’s other works
