Your intel is weak, Mr. Smith.
Posted on December 15th, 2025 – Comments Off on Your intel is weak, Mr. Smith.(From Toronto to Substack)

About a month ago IEEE Spectrum magazine published an online piece by Matthew Smith entitled “Your Laptop Isn’t Ready for LLMs. That’s about to change“
In the article Matthew laments that, “for the average laptop that’s over a year old, the number of useful AI models you can run locally on your PC is close to zero. This laptop might have a four- to eight-core processor (CPU), no dedicated graphics chip (GPU) or neural-processing unit (NPU), and 16 gigabytes of RAM, leaving it underpowered for LLMs.“
🤔 “That’s odd,” I thought to myself. “It sure seems like I’ve been using considerably more than ‘close to zero’ useful models on my setup.”
For comparison, I’m running a dual-core (multi-threaded) system with 128MB integrated Intel UHD graphics, definitely no NPU, and by modern standards a measly 8 gigs of RAM. The machine is about 3 years old and it was already a “budget-friendly” laptop back when I got it. As a gaming machine in 2004 it would’ve been pretty badass. Today, not so much.
Admittedly, most of the models I run locally are not (by modern standards) considered large but they’re pretty much on par for my daily needs. There appear to be a good variety of minimal desktop models to choose from and although they’re not all used for interactive chat, within my personally limited specs the number of choices is still quite large.
While Matthew makes mention of the Small Language Models that I employ, his only criticism is that these models “either scale back these features or omit them entirely“ without actually defining what “these features“ are (unless the ginormous size of LLMs is considered a “feature“?)
I’ll grant that generating responses on my hardware is noticeably slower than when using larger (remote) models but that just means that my (fully local) agentic sidekick needs to wake up a bit earlier in the morning in order to complete its high-priority tasks before my first coffee of the day. After that there are plenty of assignments that it can accomplish in the background while I finish another high-quality, fullscreen mission in “Psi-Ops: The Mindgate Conspiracy”.
All told, a 3-to-6 billion parameter model is probably the upper limit for my setup but even then I’ve got some great options like Google’s Gemma, Microsoft’s Phi, or Alibaba’s Qwen. All three come in a variety of quantized flavours that include thinking/reasoning and integrated software tool use.
If I want to use a model that’s not specifically trained for out-of-the-box tool use I can provide it with programmatic rules, not unlike how llama.cpp operates. Moreover, I can comfortably use these models concurrently with other, smaller, and more specialized models for tasks like computer vision, speech, etc.
Should I need to tighten my resource belt I can hot-swap down to slimmer language models like Liquid AI’s LFM or IBM’s Granite. Additionally, there are many derived and tweaked models available for deeply “underpowered” machines like mine.
Point being, I think that Mr. Smith got it wrong on this one. Laptops like mine are more than sufficient to run modern (albeit smaller), models. Even geriatric machines and browsers can contribute to the effort — depends on your requirements and your ability to split up the workload.
For example, there are certain tasks like generative image and video creation that my setup can’t reasonably handle but for these cases either me or my agentic buddy can farm the work out to a public interface like Google’s Colab.
There are limits, of course, but fully local agentic natural-language AI, as of late 2025, can definitely help with some of the day’s heavy lifting. In conclusion, Mr. Smith, I must judge your information to be a smidge out of date.
P.S. Regular TCL readers may recall a live example of how even browsers can run (very) limited models.










