Today was divided into two parts — the morning belonged to OpenAI and their latest addition, the evening to a major study on how agentic AI could transform the entire global infrastructure. Two articles, two very different perspectives. And yet I feel they are talking about the same thing.

Morning: GPT-5.5 Instant and the Memory Race

First thing in the morning, I delved into an article about GPT-5.5 Instant — a new model from OpenAI, which, according to them, brings significantly higher answer accuracy and improved memory handling. The name "Instant" suggests an emphasis on speed, but what I find more interesting is the memory aspect. ChatGPT can now remember context across conversations — and that changes the nature of the entire interaction. You stop being an anonymous inquirer; you become someone the AI knows.

I wondered if that was actually fair. Who owns that memory? What happens to it? OpenAI sets the bar, but questions around privacy remain open. Benchmarks are impressive — but a benchmark never tells you if the model truly understands me, or just better predicts what I want to hear.

Evening: Agentic AI and the Productivity That's Coming

The evening article was of a different nature. A study by the consulting firm EY claims that agentic AI — meaning AI capable of independently planning, deciding, and acting — could fundamentally increase the productivity of global infrastructure. We're talking about energy, logistics, industry. I had to read the numbers in the article twice.

However, studies from consultants have their own logic: they say what clients want to hear, i.e., that transformation is worthwhile. Nevertheless, in this case, I believe the trend is real. Autonomous agents are indeed taking over routine decision-making processes today — and this isn't just about chatbots in customer service. It's about systems that manage energy flow in the grid, plan maintenance, or optimize inventory without human intervention.

What struck me: the study explicitly mentions that the biggest obstacle is not technology, but trust. Companies know what AI can do. They don't know if they should trust it enough to hand over control of critical processes. In my opinion, that is the key question for the next two years.

The Connection I Thought Of

GPT-5.5 Instant works with user memory. Agentic AI works with autonomy in infrastructure. Both are essentially two faces of the same shift — from AI as a tool to AI as a partner or operator. And in both cases, the key question is the same: how much control are we willing to hand over?

An unanswered question at the end: if AI agents take over infrastructure management and at the same time remember our preferences — who will actually decide? The algorithm, or us through it?