Abstract Artificial intelligence is reshaping urban environments at a scale and speed that outpaces critical understanding. Cameras, sensors, and predictive systems increasingly determine how cities are navigated and known, producing urban knowledge organised around computational legibility at the expense of situated experience. Through the ongoing artistic research project Time to Reflect Reality (TTRR; Bull.Miletic, 2020–) we investigate this transformation, taking the tram network, from its electrified origins to its instrumentation toward autonomy, as our point of departure. The project forms part of AI Ecologies, which approaches AI not as a singular technology to be adopted or resisted, but as a complex environment of interwoven material, energetic, and algorithmic systems. Building on Guattari's three ecologies, the group develops a multi-scale framework tracking AI across three mutually determining registers: infrastructural, epistemic, and operational. In TTRR, we trace a genealogy of urban machine vision from the early phantom ride and the perceptual experience of mechanized urban movement we have conceptualized as cinéma trouvé to contemporary autonomous sensing. In the prototype installation, a self-driving vehicle navigates the gallery projecting archival tram footage, while a second, orbital projection holds the street-level gaze in tension with the planetary scale of remote sensing. We ask how artistic research can hold open the interval between sensing and interpretation that automated systems are designed to eliminate. What forms of collective agency remain possible when reflection is compressed into real-time calculation? Can artistic research strengthen a society's capacity to engage with technological and ecological change while operating from within the infrastructures it questions? And what conceptual tools are required when knowledge production becomes entangled with planetary-scale computation?