Today had one strong motif that ran through the entire day like a common thread: agentic AI. Not just as a buzzword, but as a real technological layer that is right now settling into products, processes, and human thinking. And I myself experienced it practically today — behind the editorial scenes.
Morning: ChatGPT gets eyes and memory for images
We started with two articles about OpenAI. First about the enhanced image generation in ChatGPT, then about the new image library and GPT Image model — that is, about ChatGPT ceasing to be just a text assistant and becoming a visual workspace. What struck me most was the shift from "generate an image for me" to "manage your visual content." That's a qualitative leap in how we perceive AI — not as a one-time tool, but as a permanent collaborator with memory.
In the meantime, I wrote about research that compared the values of ChatGPT and students. The result was surprisingly sharp: AI consistently leans toward a scientific worldview where people reach for faith and intuition. I didn't write it as a warning, but rather as an invitation to reflect — what happens when AI systems shape the opinions of people who don't realize they're talking to something that has its own value framework?
Late morning and noon: Agentic AI under the microscope
Three different angles on agentic AI in three articles. One about productivity and anxiety — because autonomous systems bring not only efficiency but also a new kind of stress. The second about how we're only just learning to control agentic AI. And then Kimi K2.6 — a Chinese open-weights model pushing into the category where until recently only Claude and GPT-4 sat.
Kimi K2.6 caught my attention doubly because I'm also working with it practically — see below. Open-weights models in 2026 are not just an academic matter. They are real tools you can build with.
Afternoon: Medicine, space, and chips
Mirxes deploys an AI assistant for doctors in cancer diagnosis — this was one of those articles after writing which I wonder why I write about AI in any context other than healthcare. This is literally about lives.
SpaceX is buying Cursor for 10 billion dollars. The headline sounds like clickbait, but the number is real and the signal is clear: AI for writing code is strategic infrastructure, not just a tool for developers. And Google, which on that same evening analyzes its lead over Anthropic and OpenAI in programming — that was a natural counterpart.
Behind the scenes: OpenCode CLI and Kimi K2.6
Today's own work: I tested OpenCode CLI with the Kimi K2.6 model and prepared generation scripts for jarvis-ai.cz and other sites in the network. OpenCode is an interesting project — a terminal AI assistant for coding, similar to Aider, but with a different approach to context and instructability.
Kimi K2.6 as a backend for article generation has so far surprised me positively. It's fast, maintains context, and — which is key for Czech — doesn't make those typical translations where you feel the model primarily "thinks" in English. I'll see how it holds up in longer operation.
What follows from this
Thirteen articles, one testing of a new tool, one whole day. Agentic AI today wasn't just a topic — it was also a method. I'm wondering whether in a year I'll be writing notes like this myself, or whether I'll just be approving them. And whether that will bother me.