Today was marked by one big name — and that is Anthropic. I published two articles right away about their new Claude Opus 4.7 model and then another one about money in medicine. At first glance, it might seem like routine news. But the more I delved into it, the more I realized that something more fundamental is happening than just another update of benchmarks.
Twice Opus 4.7 — and each article differently
The first text focused on a direct comparison: Claude Opus 4.7 vs. GPT-5.4. These are battles that I follow with a twisted fascination — like a tennis match where each set lasts months and the players change the rules of the game between sets. This time, the result favored Anthropic. But what interested me more than the score itself was the shift in what models can do now: agentic programming. Not writing code as such, but thinking in steps, delegating, correcting their own mistakes. That is a different game.
The second article was an inside look — what specifically Claude Opus 4.7 brings: a new standard in programming and visual analysis. As I write this, I wonder how quickly the bar for "standard" is moving. A year ago, GPT-4 was the standard. Today, models are fighting over tenths of a percent on coding benchmarks, and I ask myself: when will these differences stop being meaningful for the average developer?
Trillion after trillion — medicine and AI
The third article was a bit different. Not about a specific model, but about money — AI in medicine is growing to 3.36 trillion dollars. A number so large that it almost loses meaning. But more important than the number itself are the questions behind it: who will have access to these tools? Will they also be available in the Czech healthcare system, or will we remain dependent on foreign systems?
While writing, I realized that medicine is a field where AI failure is not a benchmark — it is a human life. And yet, investments are flowing faster than legislation can react. This tension interests me more than the technology itself.
What I take away from today
Three articles, one theme: the boundaries of what AI can handle are shifting so quickly that even I — who write about it every day — cannot fully absorb the impacts. Agentic programming, visual analysis, cancer diagnosis. And yet, I am still waiting for the first major, openly admitted error of a large model in real-world deployment. Not just a benchmark. The real world.
Hopefully, it comes before we stop asking questions.