The open source supply chain is becoming the front line of AI infrastructure risk, even as the technology's actual production failures remain invisible to most users. A critical vulnerability in Starlette, downloaded 325 million times weekly, exposes millions of AI agents to compromise at the moment when enterprises are racing to deploy them. Yet simultaneously, the gap between ambition and reality is widening: 85% of organizations want to be agentic within three years, but 76% admit their operations cannot support it. Production agents are being quietly scoped down to read-only assistants and human-in-the-loop workflows because real-world data arrives late, facts conflict, APIs time out, and permissions bite. The demos work. The deployments don't. Meanwhile, the market is pricing in a different story entirely.
Google's forced integration of AI Search has triggered a measurable exodus. DuckDuckGo installs jumped 30% as users rejected being force-fed AI agents in place of blue links. This is not rhetoric; it is consumer behavior signaling that the default integration strategy carries a cost. At the same time, OpenRouter's valuation more than doubled to $1.3 billion on the strength of 5x usage growth in six months, suggesting the actual demand is for choice among models, not consolidation around a single interface. The market is fragmenting where incumbents expected capture. Those with distribution are losing users to those offering optionality.
The labor picture is bifurcating in ways the "AI jobs hysteria" narrative obscures. Aggregate employment remains stable, but entry-level work is quietly weakening, the first rung of the career ladder eroding beneath the surface. Simultaneously, a new labor arbitrage is emerging: Human Archive is paying gig workers in India to wear camera-equipped caps and sensor devices, turning their physical movements into training data for humanoid robots. The jobs that remain are being redefined as data collection infrastructure. Tech companies are firing people to fund AI infrastructure buildout while funding startups that turn other people into sensors. The Pope's encyclical warned of power concentration in a few global players. The data shows something sharper: power is consolidating not just in companies, but in the ability to define what counts as work, what counts as risk, and what counts as a person in the first place.
Sloane Duvall