"Bet on token costs falling when evaluating AI efficiency tradeoffs A participant raised the valid point that using a browser agent to upload a YouTube video burns far more tokens than calling the YouTube API directly. Another speaker reframed the calculus: 'the price of intelligence is going to zero... they want to saturate the market right now so they don't care that much' about token waste, and models will eventually self-select the efficient path. When deciding whether to build a precise integration or let an agent fumble through the UI, factor in the trajectory of costs — the browser-based approach that seems wasteful today may be cost-neutral in 12–24 months while saving significant engineering time now. Source: Participant, AI in Action, AI in Action Live Session"