This Week in AI for Ridiculously Busy People

The AI industry has transitioned from a "token subsidy" era to a "token shortage" era, forcing companies to shift from flat-rate subscriptions to usage-based models. This pivot is evidenced by Uber and Walmart implementing usage caps and TSMC forecasting multi-year supply constraints. In response, the market is prioritizing token efficiency through architectural innovations like model routing, hybrid local-cloud inference, and worker-advisor agent configurations. Notable developments include Factory’s cost-cutting routing system and Microsoft’s collaboration with McKinsey, which produced a specialized model outperforming GPT-4o at a fraction of the cost. Beyond infrastructure, the sector faces intensifying policy debates as the U.S. government considers equity stakes in major AI labs amid signs of recursive self-improvement. For enterprises and solo practitioners alike, success now depends on building agent-centric systems and mastering context management to navigate rising operational costs.