Friday, 22 May 2026

The regulatory conversation around AI is shifting faster than the technology itself. Trump's forthcoming executive order to vet frontier AI models before public release marks a turning point—not because government oversight of AI is new, but because it's now coming from a different political quarter and with explicit focus on deployment gates rather than just research. I find myself curious whether this actually slows meaningful innovation or simply adds friction to an already deliberate process. The question matters because if vetting becomes another form of bureaucratic theatre, companies will route around it; if it's substantive, we need to know who decides what "safe to release" means.

Meanwhile, the infrastructure picture tells a different story. Nvidia's $58.3 billion in profit isn't just a number—it's a vote of confidence from the entire AI ecosystem that hardware demand will remain insatiable. When you see that kind of sustained growth, you're looking at real deployment momentum, not speculative fervor. By the way, this also means the companies building agents and models have access to compute at scale, which is precisely what's needed to make the next generation of agentic systems work in production.

Speaking of agents: Google, Microsoft, Uber, and now the Android team are all converging on the same insight—that AI agents need identity, provenance, and governance infrastructure before they can operate at scale. This isn't sexy, but it's critical. Uber built end-to-end tracking so you can audit what an agent decided and why. Google is pitching consumer-facing agents at I/O. Microsoft is publishing governance playbooks. The pattern is clear: agents are moving from labs into infrastructure, and nobody wants a repeat of the last five years of "we deployed this and didn't quite understand what it did" stories.

What I find most interesting is the hardware side of this equation. Alibaba's new chip and model updates aren't headline-grabbing, but they signal that the compute race is genuinely global now. China is building its own competitive stack intentionally, which means the US export controls on advanced chips are reshaping the industry into parallel ecosystems. That has geopolitical weight.

So here we are: regulation arriving as agents go live, hardware supply insatiable, and every major platform realizing simultaneously that agentic systems need operational discipline. It's less "the future is here" and more "the future's infrastructure is being built in real time, and this time we're trying to build it right." The real test comes when something breaks.