Hardware is still hard-ish

I have been close to the energy sector for the last 12+ years. One of the challenges of the space is that physical systems do not follow Moore’s law. A result of this is that the leverage to scaling hardware businesses is not as high as it is for software. This makes venture math more difficult.

While hardware marginal costs do not go to zero, the result of the effort can create a reasonably effective moat (unless China decides to target your sector, a post for another time). The places where companies have been successful is utilizing software and the Moore’s law-type effects to create physical things. It is on this software/hardware frontier that AI has been an incredible leverage point. I have seen this in my own work and am eager to see it permeating the entire space.

Like many curious minds, I have experimented with tons of different AIs. Claude code, Cursor/Windsurf, Gemini (my nieces and nephews love nano banana) and all of the interesting open source models are always worth poking around as new versions come out. Replit has allowed me to create random app ideas and push websites to production.

But where AI has directly impacted the hardware space is in applying the intelligence to some of the complexities of physical systems. I haven’t yet seen AI completely understand physical space but it is great leverage in building things yourself. It can’t connect the proper wires, but it knows and can find the calibration settings and sensitivities of sensors. De-bugging used to require reads into manuals, address mapping and infinite other rabbit holes that most engineers are familiar with. My experience has been incredibly liberating and empowering where the slog used to be an endless exercise in frustration. I have seen my dad troubleshoot his electric golf cart battery management system written in some 25 year old scheme that lives in the depths of forums and my cousin get functioning python code for an actuation system in a day. Both of them are mechanically capable and handy with a wrench, but AI enabled them to build and solve their challenges.

AI is also a great directional thought partner for things as basic as part selection. Do I buy this chip or that? Feed AI two different data sheets and it can immediately parse them and get you in the right direction. They still show up from Mouser and have to be connected but you know that once you plug it in, the AI will be there to get the proper drivers and protocols.

I don’t yet have a refined workflow I could point and say the best way for each person to use AI, but I know it makes my life much easier and multiplies progress. Software engineers for sure still have it better than hardware engineers, as I am reminded when software only tasks glide by smoothly. Devin and Codex and other assistants always create their work environment but they can’t quite recreate the basement environment, yet.