Gemini can just consume the device documents. There's an incentive for device makers to publish this content.

There had always been some incentive for manufactures to publish device documentation, and yet it has often been quite lacking either in quality or overall existence. I doubt LLM/agents being the readers will change that at all. What I expect AI scraping and using without credit will impact is people publishing their own unofficial help and guidance, and the affect there is likely to be negative. It won't stop all of them, but enough to be noticeable. Another possible negative is the manufactures documentation being AI generated without sufficient review, so possibly more erroneous than before, or intentionally not producing full documentation at all and expecting AI to fill the gap (MS seems to be heading this way: pushing "ask copilot" all over Azure instead of links direct to good reference material). All this would add up to a situation that is somewhere between "a little worse than pre-AI" and "an absolute shit show".