"We're just giving you the tools." The lie that scales.

Meta's announcement that it open-sourced Muse — a framework for building custom AI gadgets from cheap hardware like ESP32 boards and HDMI sticks — reads as a gift to hobbyists and makers. The real move is structural, not generous. This is how Meta learned to build moats that look like democratization.

In 2011, Meta's Open Compute Project released hardware designs and server specifications as open-source infrastructure. The framing was identical: level the playing field, let anyone build efficient data centers, break the vendor lock-in of proprietary equipment. What actually happened was different. By establishing Meta-compatible specifications as the de facto standard, the company became the inescapable reference implementation. Data center operators gained the freedom to source components independently and gained the requirement to stay compatible with Meta's architecture. The openness was real. The trap was structural.

Muse follows the same pattern with a different substrate. A hobbyist can absolutely load Meta's AI agent onto an old E Ink display or a secondhand touchscreen. The code is open. The hardware is cheap. But the moment that device needs to be useful. Needs to understand context, sync data, interoperate with other Muse devices. It depends on Meta's agent runtime. You can modify the interface. You cannot opt out of the layer that makes it work. That layer is where the data lives.

The tension dissolves once you see it's not about whether this is 'genuine' openness. It's about which part of the system Meta doesn't need to control because the architecture controls it for them. Open Compute didn't require data centers to use Meta's servers. It required them to be compatible with Meta's way of thinking about servers. Muse doesn't require your DIY device to report to Meta. It requires your device to speak Meta's language to be interoperable. Scale this across thousands of maker projects and you've built distribution through user autonomy.

The real danger isn't that hobbyists lose something. It's that as DIY Muse devices proliferate. Each one a small edge node reporting telemetry, training the agent, normalizing the dependency. The ecosystem quietly consolidates around a single reference implementation. Meta owns the layer nobody thinks to question because it's the only layer that makes everything else work. That's how infrastructure becomes inevitability.