Can AI write your code the same way it writes your app? That question has a tentative yes for software. For physical devices, it's a different story.
Schematik wants to be the "Cursor for hardware" — a vibe coding tool that lets anyone program robots, embedded systems, and IoT devices through natural language. According to Wired, Anthropic has expressed interest in the project. The pitch is seductive: democratize hardware programming the way Cursor democratized software development. The thesis sounds logical until you examine what hardware actually requires.
The core problem is not intelligence. It is physics. Software vibe coding works because the cost of being wrong is low. When Cursor generates buggy code, you see the error instantly, fix it, and try again. You can iterate hundreds of times in an hour. Hardware inverts this equation entirely. When a vibe-coded robot arm malfunctions, the error is not an exception message. It is a shattered servo motor, a broken wrist, or worse. The feedback loop that makes vibe coding magical — write, see, correct — stretches from milliseconds in software to hours or days in hardware. Components wear. Prototypes break. You cannot ctrl-Z a melted microcontroller.
This does not mean the vision is wrong. It means the starting conditions are radically different. Software is pure information. Hardware is information constrained by metallurgy, tolerances, thermal dynamics, and safety margins that no AI model can fully simulate. A vibration algorithm that works in simulation can destroy a real motor because the simulation never modeled bearing friction at 3,000 RPM. These edge cases are not edge cases. They are the job.
Proponents will argue that AI is improving fast, that safety guardrails can be baked in, that the democratization potential justifies the risk. They are not wrong. If Schematik succeeds, it could do for embedded systems what no-code tools did for web apps — open the field to people who understand problems but not registers and timers. That is worth pursuing.
But the gap between "works for a React component" and "works for a drone flight controller" is not a sprint. It is a category shift. The teams that will actually ship vibe-coded hardware in the near term are the ones that treat the physical world with the respect it demands — using AI to assist engineers who already know what can go wrong, not to replace that knowledge entirely. Cursor made developers faster. Schematik wants to make hardware accessible. One of those goals is achievable today. The other is a moonshot wearing a very compelling pitch deck.