The criteria for usefulness are variable, like individual needs; in a way, it's something personal in terms of the intended use, which can then have implications for the community.

But in general, it's like having applications similar to those used for other payment methods. I think it would be exceptional in the case of LN, without excluding regular transactions.

AI is currently helping people a lot, but not about the future. Any information can be obtained with the help of AI at any time if needed. It gives people a good idea about their business, how to be profitable and helps any person in their work and makes any difficult task easier. AI is quite interesting to me, but it will help people more in the future so it can gain more popularity.
Any information about Bitcoin can also be obtained from AI, so I think it is suitable to help more in the future.

AIs are evolving and are learning from feedback. I asked an AI to help me edit my PDF file, it couldn't, it kept on OCRing the file, until I gave it a precise instructions, it then apologized for the previous mistake but told me it couldn't. I then asked it to look for a way, which it then took 3 steps to finally get me my desired result.

Most AI models can learn from feedbacks and are also evolving. In the future, it will be much more reliable in different fields and aspects of life. Before using an AI to optimize and speed up processes, you should have a considerable amount of knowledge in that process so as to easily id whatever mistake. And besides, there is often a warning on almost all AI models I've used, which is: AI can make mistakes, make sure to verify the information it gives. If we follow this warning, AI can most definitely be a very useful tool in every aspect of our lives. And I'll also advise to learn and understand prompting too, cause this can also refine the type of output/answer/data you get.

---

Concerning these two links, the first link: with or without an AI, the coldcard vulnerability exists either way; sooner or later, an ambitious hacker would've figured it out and done something similar with or without an AI. AI didn't create that vulnerability, humans did. And if they had used an AI to figure it out first, then we would've been applauding and not criticizing AIs, I guess.

And the second link was just blown out of proportion, it's not like an unthinkable thing happened. The sandbox had a loophole which was exploited by an AI model that is trained to exploit loopholes. It's like you lit a match and got surprised it created a fire.

Sure, local AI is probably safer, in theory. There is still an attack vector though, and anyone that breaks into your computer would then have control of anything that AI agent can do, which includes creating crypto wallets and signing crypto transactions. Youre adding an extra layer of software that you have to trust. Bitcoin has traditionally favored the boring, slow path of fewer software layers and private keys stored offline, boring approaches which tend to fail less often when there is money on the line.

I see more potential for AI use in crypto in analyzing protocols, reviewing code, threat modeling etc. than for key management or transaction signing.