"Artificial intelligence (AI) increasingly helps communities make decisions related to healthcare, agriculture, education, and climate adaptation, yet most AI systems are built for people in connected, urban settings who speak dominant languages such as English. Indigenous and rural communities are often left out of the data, the model design, and the subsequent decisions that follow: a problem AI has not created, but has intensified. Using the Andes of Ecuador as a case in point, this commentary argues that this is a cyclical problem that is older than AI: where outside actors introduce a technology and decide how it will work without considering diverse cultures of people, often rural or Indigenous, who will use it. Unlike infrastructure projects such as a road or clinic which are commonly built in rural communities without prior consultation, commercial AI systems similarly are currently impacting communities but are controlled by distant companies, leaving community members with no clear local mechanism to contest or correct culturally inappropriate model outputs. Drawing on examples from Kichwa communities, we show how language, timing, and collective decision-making rarely fit the way most AI systems are designed, leading to advice that is misread, mistimed, or simply ignored. We then outline what a negotiated, culturally sensitive approach could look like: outputs checked by local community hierarchy before they reach the broader community, data used only with permission, and voice notes or images accepted as input, not only typed text. This is a concept, offered as a starting point to open a broader discussion in collaboration with developers and policymakers who want AI to serve these communities rather than sideline them. What is at stake is not only access to new tools, but who gets to decide how knowledge is used, and who can refuse that use altogether." (Abstract)