It is not the first time that a word becomes fashionable in the technology sector. It has happened with IoT, Big data, Blockchain and even 5G. In English they call it buzzword and refers to those terms that are repeated and repeated until they almost lose their meaning. It has happened with AI and, now that we have overcome that first stage, it was time to give it a surname. The chosen one is agentic AI and, suddenly everything is agentic AI.
I experienced it a couple of weeks ago in the Qualcomm Snapdragon Summit. During the different conferences, the most repeated words were “agents” and “agentic”. The problem is that they didn’t show any real products that actually fit this definition. They are not the only ones, there is a whole wave of companies that already call agentic AI literally anything that has minimal automation.
Agents all the time everywhere
Agentic AI It was going to be a revolution in 2025but reality ended lower the smoke to the gurus of the sector. With this I don’t mean that everything is a hoax, AI agents are very real and they are already here. We can try them If we have the ChatGPT Plus plan. At the development level, Anthropic allows create scheduling agents with Claude and Google with Gemini. Other platforms like Salesforce offer their own custom AI agents for specific sectors such as public or industrial. They are improving a lot, but the reality is that AI agents They are still very green as has been demonstrated in enough tests.
There is no real product, not even one that they are developing, everything is part of a dream, one in which AI agents are even in the soup.
Being cautious and waiting for technology to develop does not suit many companies. Returning to the case of Qualcomm, in the “The Ecosystem of You” conferenceits CEO Cristiano Amon drew us a future in which “the agent” does everything for us, absolutely everything: “The agent will understand our world and will be helping us, anticipating every need.” The problem is that everything he showed It was simply a demo. There is no real product, not even one that they are developing, everything is part of a dream, one in which AI agents are even in the soup.
What is agentic AI
It is also known as agentive AI, agential AI or simply “AI agents”. Google defines it as “an advanced form of artificial intelligence focused on autonomous decision-making and action.” For NVIDIA is an AI that “uses sophisticated reasoning and iterative planning to autonomously solve complex, multi-step problems.” For amazon It is “an autonomous system that can act independently to achieve predetermined goals” and they add that, unlike generative AI, agential AI “is proactive and can perform complex tasks without constant human supervision.”
It seems pretty clear, generative AI responds to one request at a time, while agentic AI can achieve more complex goals, making decisions autonomously. An AI agent must be able to collect information, use tools and solve problems to achieve the objective we have given it.
They call it agentic AI because “AI with slight automation” doesn’t sound so good
In another of the Snapdragon Summit conferences they showed us several products that were real, one of them is Page.aian AI assistant that works locally on mobile. During the intervention, the presenteror stopped repeating that the app had agentic functionswhen the most they showed was how the AI was capable of organizing a barbecue: it created an event on the calendar and then invited a friend. What caught my attention is that the creator of the app did not use the word, but was the presenter.
The reality is that many of the use cases presented as agents are, at best, a kind of IFTTT on steroids. In this CNBC articlethe head of AI at the consulting firm EY assured that “Many in the market want to take advantage of it. We have witnessed an incredible change of image of everything related to generative AI, which is now presented as agentic AI.”
“Agent washing”: when products that are sold as agents are actually products that already existed.
At the beginning of the year, Gartner surveyed more than 3,000 companies who promote AI agents and discovered a trend they call “agent washing.” That is to say, many products sold as agents are actually products that already existed. Gartner estimates that of the 3,000 companies, only 130 sell real AI agents. “Most agency AI proposals lack significant value or return on investment,” said analyst Anushree Verma. The firm predicts that more than 40% of agentic AI projects will be canceled before the end of 2027.
Why so much hype?
In May of this year, a survey of senior American executives revealed that 88% of companies had planned to increase their AI budget before the arrival of agents. Most respondents believed agentic AI was going to change workplaces more than the internet did, and nearly half were worried about competitors adopting AI agents before them.
The fear of being left behind has encouraged many companies to jump into the pool without fully understanding what agentic AI is. It makes sense that they want to hype it up and even “cheat” by calling agents who really are not: they are investing a lot in this and they need it to turn out well.
Image | Gemini
GIPHY App Key not set. Please check settings