The blogger who made the only insightful point about AI is that all your efforts are futile in the face of large-scale model iterations. Take OpenClaw, which has been trending recently. Is it impressive? Yes. Can you use it? No. Due to limitations in context, memory, and cost, it can only be a "lobster pet," not Jarvis. However, as large-scale models iterate step by step, these problems will be solved in a short time. So, while OpenClaw may not be powerful now, it will be very powerful in the future—that's its significance. I laugh when I see many people complaining about OpenClaw's high resource consumption. Of course it's powerful! It's just randomly storing data; does it have any context? Does it have to try things over and over again? But in reality, most of the work requires something like an MCP (Multi-Purpose Component Model), and only after the MCP has undergone a major shakeout will the true MVP (Most Valuable Product) emerge. As for having it do the work for you, it's really more cost-effective to thoroughly try all the mainstream large-scale models and then choose the one that suits your needs, because the latter can indeed solve your problem. But playing around is still fun. Who doesn't want a smart electronic pet?