Just a year ago, ChatgPT-4 seemed like a piece of magical technology. It was impressive. I could write essays, write code, generate stories and even offer financial advice with almost human fluidity. But here we are, in 2025, and what recently flew our heads already begins to feel a bit rudimentary.
The reason for this is not that language models have stopped improving, but the way we use them is changing radically. Openai has just launched Deep Research, a new functionality that allows Chatgpt to act as an autonomous agent capable of planning and executing complex searches in multiple steps. This is not just an update, it is an indication that the era of the LLM (Large Language Model), as passive conversation attendees, are coming to an end.
Until now, interacting with an artificial intelligence (AI) as chatgpt was a round trip: we asked questions, we got answers and, if the information was not enough, we adjusted our instructions and asked again. But Deep Research changes this dynamic. Now, the model not only generates text, but also reason about the search process, goes back when necessary and evaluates sources to extract the best possible information. In other words, what previously required trial and error sessions with the model, now happens automatically and optimizedly. Now the LLM begin to look more like a complete team of attendees working in the background.
How fast is being progressed?
This transformation is not happening only in Openai. Google, Deepmind and Anthropic have been experimenting with similar agents. In December, Google presented its own prototype, Project Mariner, which follows a logic similar to the Deep Research of OpenAi.
The really interesting thing here is the speed at which these advances are happening. The models of two years ago already seem prehistoric. At this rate, how will we see the current LLMs within six months? It is very possible that, in retrospect, we seem primitive tools compared to what is coming.
If this is the year when the great language models as we know them become obsolete, it is because they will cease to be the center of the experience of AI. Instead, agents will take control, executing tasks autonomously and delegating work without human intervention.
What tasks will remain in the hands of the agents?
- Complex research: To analyze market trends, review legal documents or any research task that took us before. It is necessary to read and interpret documents, an agent will do it for us.
- Decision automation: From programming appointments to managing investments, agents can optimize processes without constant human intervention.
- Assisted creativity: If the LLMs generated text and images before, they could now plan strategies of marketing complete or even design entire products. He input Initial human will be key for agents to meet expectations, but much of the creative execution will be left to these agents.
The most striking thing is that these agents will not be reserved for technology experts. Any person may have a team of digital assistants executing tasks in the background, almost as if they were virtual employees.
The evolution of AI agents raises fundamental questions: What will happen to our relationship with information when we no longer need to look for it? How will we redefine work when a significant part of our tasks can be delegated to a machine? Will we adapt faster than we imagine?
If we have learned something in the last two years, it is that artificial intelligence does not expect us to get up to date. The real question is not whether the LLMs will be obsolete, but how soon we will realize that they are already.
#extensive #language #models #obsolete #year
