Wednesday, 27 May 2026

The past few months have clarified something I've been thinking about since the release of Claude Code and OpenAI's agentic systems: we're not watching AI adoption flatten anymore. We're watching it fracture. The broad strokes tell one story. The details tell another.

JPMorgan's latest analysis captures this tension perfectly. Agentic AI deployment among large enterprises has more than doubled, which is significant. But overall AI engagement metrics are stalling. What does that mean? It means the organizations with resources and technical depth are moving aggressively into autonomous systems, while the rest of the market is stuck. The promised democratization of AI feels further away, not closer. The gap between leaders and followers is widening in real time.

This matters because it signals where real value is being created right now. It's not in chatbots or document analysis anymore. It's in systems that can independently navigate software, make decisions, and execute tasks at scale. Microsoft's Copilot Studio is pushing computer-using agents forward with better workflow control. Amazon is building serverless infrastructure specifically designed for multi-agent systems. These aren't incremental product updates. They're betting on a world where AI doesn't assist humans but completes work autonomously.

Which brings me to something worth sitting with: Claude Mythos, Anthropic's reasoning model, recently identified over ten thousand software vulnerabilities in a single security initiative. Most were high or critical severity. I find that number both reassuring and unsettling. Reassuring because it shows AI can do genuine, technical work that matters. Unsettling because if AI agents are now finding security flaws at this scale, they're also becoming sophisticated enough to potentially exploit them. The safety question isn't theoretical anymore. It's operational.

By the way, there's a reason the Vatican recently released its first major encyclical on artificial intelligence. Pope Leo XIV isn't wrong to flag autonomous weapons and governance challenges as defining questions of this era. When deployment is racing ahead of alignment, when the gap between capable systems and safe systems is growing, institutional voices need to speak up.

Google's new video-cloning technology with Gemini Omni is impressive and concerning for similar reasons. Photorealistic synthesis of speaking avatars with natural language control is genuinely novel. The implications are obvious and worth contemplating now, not after the technology spreads. By the way, that's exactly the moment we're in across multiple domains.

The real story this week isn't that AI is transforming business or that hype is justified. It's that the frontier of what's possible has moved so far ahead of what's being widely understood that we're operating in two different eras simultaneously. That's not sustainable.