Over the years, we have heard that artificial intelligence will change our method of using Phone And the computer. Instead of taping the applications and clicking the programs themselves, AI will look for our devices themselves to answer our questions. These so-called AI agents have felt out of reaching over the years-but with the latest generation phone chips, it may last.
Snapdragon Summit in Maui 2025, I sat down with Qualcomm AI Chief Durga Malladi, who explained how this company’s newly announced Snapdragon 8 Elite Jenners 5 The chip will give strength what the first phones can be operated by large language models to the true AI agent performance.
“I believe we are among the more personalized assistants,” said Malladi. He mentions two LLMs, Page.ai And SomethingIt was said that both could be used as a personal assistant during the summit.
“It is actually the local document and key to the local you have, which all the things stored on the device and simply tap it and got some additional information,” mentioned Malladi.
The right LLMS and the hardware lined are part of the way for AI agents. But another story found by people to use them.
Are people ready to use AI agents?
Last year, first laptops packing on Qualcomm Snapdragon X Elite Chips for PCs were debuted. Despite the dynamically no separate UI, they were still rudiments of generator AI interfaces with ChatGPT-national prompts. Malladi gave an example: Asking his computer, Qualcomm CEO Cristiano Amin is asking what the last email sent to. She is now accustomed to using prompts to control her computer.
“I think people will become accustomed to this kind of behavior, and once you start to get used to it will never come back from there,” said Malladi.
However, it will take a lot of familiarity to understand how people can use AI agents and become accustomed to them.
Malladi said about AI, “As people get more accustomed to,” You realize how much time we actually spend a lot of worldly work – it’s a confusion from the original work that you have in your mind. ”
AI agents can help them become the power users of their device without deepening their settings. We discovered the new features that we didn’t know about our phone, Malladi mentioned – these are complicated devices that are not surprising that we do not know how we will use them with full possibilities. Malladi said that simply by asking AI agents to do so can simplify access to more capabilities.
Vision Clean: Automate the worldly for moments that induce new problems in the human error and automat the cover. Nevertheless the generator had its own struggle with the appropriateness of AI, Chatzipt with non -existent book Hallucinating on a summer reading list The court cases never happened In Google’s AI Overviews Tells people to eat stonesThe However, Malladi saw the scene that they had problems with LLMS two years ago; He said that has been maintained since then – both at the chip level and by Qualcomm at the LLM level.
We have to wait and see if the Malli is correct about the AI agents. The technology is already here, appearing on phones and laptops using Qualcomm chips. And if it is this year or next, the possible impact of AI agents is significant: how do they use machines that allow us to contact us and contribute to the world.
“These schools and college-minded kids who are growing up in this agent AI environment, I think they will ever know what it was before,” Malldi was difficult.
There is no other way to prepare people for AI agents without giving the technology packing devices. And to the end, the phones may be able to support them in the end, partially due to progress on Qualcomm’s mobile chips.
Qualcomm Chips: Advanced Silicon improves AI agents
Two years ago Snapdragon 8 Jenner 3 Can handle 15 tokens per second, processing a metric AI request for speed. The new Snapdragon 8 Elite Jenner can handle 220 token per second. In short, it should be allowed to process the AI assistant to quickly process the requests and further data consultation for better reaction.
Dr. Vinesh Sukumar, senior director of Qualcomm AI Product Management, said, “Last year there was the difference between 5 tokens this year, there are several seconds of 220 tokens.” “Wee affects the Human-Messine interface by exceeding the second
In the past two years, Qualcomm has switched its mobile chip to ARM-based Cryo to its own Orion Central Processing Units and increased graphical capacity and neural performances. Even last year’s Snapdragon 8 Elite improvement – 37% better CPU performance – enable AI capacity that “is actually a permissible difference that we will see,” mentioned Malladi.
“We’ll always see on-power that will actually help you,” said Malladi.
The company’s new Class of Orion CPU debuted for the PCS for its Snapdragon X Elite Chip, which entered the laptop PCs, which began in June 2021. The latest Snapdragon X2 Elite and X2 Elite Extreme Chips have been unveiled with the Snapdragon Summit this year, Maladi points to a new fence -trill -trill -trill. The new chip supplies 80 tops, the first X elite to double the 45 tops received by the first X elite.
These AI capabilities, combined with the improvement of the performance on last year’s chip, have not yet seen an era of AI agents in the PCS. Malladi admits that PC laptops with the old X elite have the same appearance and feeling of the conventional computers despite the AI capacity. But despite the impressive performance and battery life, at the end of the day, the UI has not changed so much. The goal of Qualcomm is to improve the performance of X2 Elite and Elite Extreme and to the top of the top organizations to jump to the AI agent interface.
In other words, the hardware is not a road to create AI agents on PC and mobile devices – it now depends on the manufacturers to integrate them. In these very first days, there are rarely Sens Camertis about what AI agents will do. Device manufacturers are still determining how to take advantage of Qualcomm’s chips. There are still a lot in the air about how many and key size LLMs will live on a device (or connected through the cloud) how to actually see and interact with AI agents.
Despite this uncertainty, Malladi feels that the AI agent’s gun has been dismissed.
“Are we waiting for anything? No, probably no. Agent AI is here, but you have to see it to believe it,” said Malladi. “But if you see it you would have loved, oh yes, I got it right now.”
First AI Agent PC published in Snapdragon Summit
One day after the new mobile and PC chips was unveiled in his Snapdragon Summit 2021, Cristiano Aman sat down with the CEO of Cristiano Aman Humain. In the morning, the sunlight spreads over the western mountains, both men concentrated on silver laptop between them: Humine’s first PC They describe an AI agent interface with which.
That interface was more like a browser window that has eight prompt templates from the specified (stock advice) to General (asking “something” like a chatzPT). In other words, it looks like more generator AI assistants that we are already known and less broad AI agent Vision Coalcom and pitching at the previous summit. The goal of Qualcomm is a simple: a centralized prompt window where people can ask questions and search for applications across applications and personal data to provide answers. (Publish: CNET’s main organization, Gif Davis, filed a lawsuit against the Openai in April, alleged that the company had violated the Copyright in AI systems in training and operating.)
If nothing else is the victim of the programs and the people will practice to start getting acquainted with the agent-forward interface instead of answering themselves. If they are typing questions on their PC keyboard or vowing them on the phone, people need to adjust to a new example of using devices.
AI agents will be used separately between PC and mobile platforms, Malldi predicts.
People will ask questions over the phone and go through one or more LLMs to get the AI agent answers. However, power limitations on a mobile device will limit its abilities to integer-based processing, basically since it will run on the device. LLMs are limited to relatively low parameters in single digits (1-, 3-, 5- or 7-billion parameter models are the most common, which means less accuracy).
On PCs, agents will focus on cloud-based models, which turn off the device. They will probably be more applicable for enterprise applications and floating-point-based models instead of integer-based models. In the end, they will be used to increase productivity for work such as creating a code line.
“Half the time you wrote the same code that you wrote earlier and then you introduced some bugs unnecessarily,” said Malladi. AI can produce that repeat code so that programmers can concentrate on problems that can solve the novel’s methods and problems.