Luca Bertelli

Are we approaching the end of the hardware as a service era? The pattern I see: you connect your coding agent to the firmware you study it, replicate it like people are doing with adobe/games. Encryption and other security features might protect in using existing services, but users can always flash the device with new os and have their coding agents rebuild the services from a blank canvas. It takes only one person doing it and sharing the method in an md even, but even this is not essential, just makes it more efficient. > I ve had friends who jailbroke 8 sleep, to get metrics without sub >@ilbert_luca is trying this with Alexa, just to give her an openclaw brain, so she’s not stuck to 2013. >I saw a repo doing this for whoop And I think there’s many more examples. >the posts going viral with spyware on Samsung TVs where people are using zero day vulnerabilities found with coding agents to “look inside” The exciting thing about this: You can bring back to life wearables where the tech is amazing but companies discontinued them (my beloved Oxa, your time to work again is coming). Also, one off purchase devices usually are open with their data. I’ve been playing around with my concept 2 rower, which connects with whoop for Hr and it’s super fun. What might prevent this: 1. Community features - you might continue paying for whoop to mog your friends with your sleep scores. I don’t think this will be the case as new communities will emerge, and if they are good enough people will connect with friends there. 2. services requiring heavy compute and proprietary algorithms- offline processing of sleep data, algorithms that where built/verified by/with big volumes of population data might be more accurate or efficient to run on bigger servers. I think this will be more important for medical devices rather than leisure ones. 3. Devices that give you access to ecosystems, like Google, Apple or Amazon assistants. But with the agentic web evolving alternative ecosystems will emerge and new assistants will access them. And if it’s not entirely possible today (which I believe is to a big extent) it will be completely possible in the following months. Curious to see if this to the limit will extend to robots, as robot learning improves and hardware is depployed. Will this be the equivalent of open weight models to the big labs today? I guess the cycle comes back, and the way to make money from hardware is owning, renting it or leasing it.