Tuesday, 9 June 2026

Apple just showed everyone what happens when you stop pretending Siri is useful and actually rebuild it from the ground up. The new version, powered by Apple Intelligence and built in partnership with Google and Nvidia, is genuinely conversational and woven into the device ecosystem in ways that finally make an on-device assistant feel essential rather than ornamental. This matters less because Siri is now "better" — though it is — and more because Apple's execution suggests the era of cosmetic AI features is ending. When the world's most valuable company commits engineering resources to make an assistant actually understand context and intent, it signals that assistants aren't a marketing checkbox anymore.

By the way, what's happening in parallel is equally striking. Anthropic is quietly pushing AI agents into biology, while MetaMask has launched Agent Wallet to let AI systems manage crypto transactions autonomously across multiple blockchains. These aren't flashy consumer products, but they're the plumbing. We're moving from "AI assists humans" to "AI acts on behalf of humans in constrained domains." The security implications are real — MetaMask built in safeguards for a reason — but the trend is unmistakable. Within months, we'll have AI making decisions about drug compounds, financial swaps, and device configurations without waiting for human approval each time.

Here's where I find the disconnect genuinely uncomfortable. Microsoft just signed a massive deal to deploy Copilot across 505,000 NHS clinicians in the UK. That's not a pilot; that's systemic adoption of AI in life-or-death decisions. Yet the safety conversation happening in parallel — led by people like Stuart Russell and organizations like Anthropic — is pointing out that we don't actually have consensus on how to govern these risks, and that private companies building these systems can't be trusted to self-regulate at the speed required. Oliver Meier's argument that the Manhattan Project needed government coordination, not just lab directors making promises, lands harder when you realize Apple, Microsoft, and Google are simultaneously shipping systems into billions of devices and critical infrastructure with minimal external oversight.

I'm not arguing these products shouldn't exist. I'm saying the gap between capability deployment and safety governance is widening visibly, and pretending it's narrowing would be naive. The question isn't whether AI agents and smarter assistants are coming — they're already here, being handed to clinicians and wallet holders now. The question is whether we'll establish meaningful oversight before the next generation makes decisions we can't easily reverse.