RTX Spark: Local AI Baby Steps and a Few Small Victories ⭐

Weeks into my RTX Spark experience, I remain confused that Microsoft and Nvidia released this platform now and in its current state. This platform is not ready in every sense of the word: Laptops are the wrong form factor for local AI agents, the devices are heavy, hot, and loud, the memory isn't really unified, there are no features in Windows 11 that take advantage of the hardware or let you manage it in any way, and local AI is complex, slow, and unlikely to meet anyone's needs, especially at these price points.

It's been amusing to see that other reviewers also struggle with these issues. Less amusing is how they handle the strange experience of having their expertise challenged by technology they (and I) do not fully understand. In the span of two days, the same PC World reviewer published two articles that do not provide a single valid reason to get one of these laptops or say anything meaningful at all. And to be clear, that's not a criticism, it's a commiseration. I get it, we're all struggling to make sense of this platform.

And yet.

Not at my experiences with RTX Spark computers have been negative. No, they make no sense for over 99.99 percent of the population today. They're here well before the local AI and agents they purport to exist for even make sense. And they're so expensive, most customers will need a mortgage-sized loan to buy one. But at least running local AI is incredibly slow and tedious, am I right?

Well, yes, that is sadly correct. But also a joke. As noted, it's not all bad. The hardware is high-end, and of premium quality, and each PC maker--at least of the three I've tried so far, HP, Lenovo, and ASUS--has put their respective design stamps on their laptops to good effect. Most portable workstations are big, bulky, and unexceptional looking. RTX Spark computers are portable workstations with style. Let's at least admit that.
Local AI just isn't as familiar and it just isn't as capable
I had only dabbled in local AI before my RTX Spark experiences, though I should qualify that: I use Copilot+ PCs extensively and I do use the local AI features in this platform--Click to Do and the photo manipulation capabilities in Paint and Photos, especially, but also others--daily. What I mean here is that we have well-established AI assistants, as we're now calling them--things like ChatGPT, Claude, and Gemini--and the local AI equivalents, which we need a good name for, work similarly but require a lot more work and knowledge, and are less familiar. I had only dabbled in these apps--LM Studio being the most common--before RTX Spark. The key exception is my using of Clairvoyance (with Anthropic Claude Code) to finish my WinUIpad coding project earlier this year.

But with RTX Spark, I've dived into the deep end of the pool before I could swim without assistance, so to speak. I've enlisted the help of friends like Brad and also Brad's boss, confusingly named Brad, the founder and owner of Stardock and author of...