Why the brain needs to be portable
Two robots can look completely different underneath: different motor controllers, different camera modules, different microcontroller boards, different wiring. If the agentic reasoning engine had to be rewritten every time it ran on a new robot, none of the reasoning, prompting, and planning work that went into building it would carry over. A hardware abstraction layer solves this by sitting between the agent and the raw hardware, presenting a consistent interface no matter which specific motors, sensors, or boards are underneath. The agent asks for move arm to position or read the front camera, and the abstraction layer figures out what that means for this particular robot's actual hardware.
This is conceptually the same idea as an operating system abstracting away the differences between hard drives so that application software doesn't need a different version for every disk model. In robotics it means the same agentic brain, complete with its skills and its decision-making logic, can be ported across different physical bodies with comparatively little rework, as long as each new body exposes the same abstracted interface underneath.