@yiliuai: One of my biggest takeaways at MIT is that the world is just a giant makeshift operation. I experienced it deeply again today. This semester I'm taking a course on AI Agents, which requires group projects (the professor said he prefers infra over app layer). Inspired by my previous project connecting an LLM to a vibrator, and because I'm interested in hardware, I originally wanted to build a protocol or middleware layer that allows AI agents to connect and control most hardware. Coincidentally, another MIT undergrad wanted to do something similar; he invited me to join after learning about my vibrator project. But after careful consideration, I abandoned this project and instead chose another more superficial app-layer project. (I don't care about grades anymore, and I couldn't find a project I was both interested in and saw promise for.) My reason for giving up on this direction I was initially very interested in: I believed it had neither commercial value nor any technical moat.

One of my biggest takeaways at MIT is that the world is just a giant makeshift operation. I experienced it deeply again today. This semester I'm taking a course on AI Agents, which requires group projects (the professor said he prefers infra over app layer). Inspired by my previous project connecting an LLM to a vibrator, and because I'm interested in hardware, I originally wanted to build a protocol or middleware layer that allows AI agents to connect and control most hardware. Coincidentally, another MIT undergrad wanted to do something similar; he invited me to join after learning about my vibrator project. But after careful consideration, I abandoned this project and instead chose another more superficial app-layer project. (I don't care about grades anymore, and I couldn't find a project I was both interested in and saw promise for.) My reason for giving up on this direction I was initially very interested in: I believed it had neither commercial value nor any technical moat. But today, at the demo day for this course, that very project won first place in the scoring by the many VC investors and judges present. At the time, my reasons for abandoning this direction were as follows: This can be broken down into two types of devices: consumer-grade and industrial-grade. For consumer-grade devices, there's no technical challenge, and connecting them doesn't add value. The vast majority of consumer electronics use open protocols. In the US, there are mature apps like Home Assistant that already connect everything available on the market. The reason I had to reverse-engineer that vibrator was because it used a proprietary BLE protocol not found in any public standard—that's actually the exception. Common consumer devices don't need an LLM to "understand the protocol"; you can just call existing libraries directly. So even if you use an agent to connect them, you're barely creating any additional value because the problem is essentially already solved. For industrial-grade devices, the situation is much more complex, and a school project simply cannot cover it. Each device has different protocols, interfaces, and control logic. You either have to invest a lot of money to buy many devices and get them working one by one, or align directly with manufacturers—both of which are impossible within a one-semester course project. At the time, I considered that perhaps as a course project, just implementing connections for four or five devices, or convincing four or five manufacturers, would be enough to pass (I didn't expect they'd get away with just making one easy-to-connect device work lol). But in the long run, I didn't think this was suitable for me to start a business either—my own hardware background and industry resources are not strong enough to get most manufacturers to adopt a protocol I defined, or at least I'm not the best person for this. Moreover, I believe that in the future, hardware manufacturers themselves will make their devices more convenient for AI agents to connect and control, so there's no need for a third-party middleware or protocol. Even if I were the right founder to orchestrate this, what specific problem does this "connection" solve? How much value does it bring, and who will pay? At that time, I discussed with that undergrad, and he said the target users are mainly developers, who might pay for it. But if that's the case, I seriously doubt developers would be willing to pay for such a thing, so I saw limited prospects. What I didn't expect was that merely getting one device to work could really be marketed as "a protocol that connects AI agents to all/general hardware" (universal hardware control) and then take first place. What they ended up doing was: on an open-source educational robotic arm (SO-ARM101), they did three things: added an interface that AI could call, added a script for automatic restart on error, and added a camera so the AI could check if it made a mistake. The underlying SDK for this robotic arm had already been packaged by Hugging Face's LeRobot team—one line of Python could call it. But they packaged this as "Universal agentic hardware control", paired it with a slightly tech-looking landing page, and fooled the VCs and judges into giving them the highest rating. This world really is a giant makeshift operation.