Agent AI is here to improve your life, you just have to believe in it

The way we use it Phones Computers will change thanks to the beginning artificial intelligence – Or at least that’s what we’ve been hearing for years. Instead of clicking through apps and clicking through programs ourselves, AI will sift through our devices to answer our questions. These so-called AI agents have felt out of reach for years – but with the latest generation of phone chips, that may finally be changing.

At the Snapdragon Summit 2025 in Maui, I sat down with Qualcomm’s head of AI, Durga Maladi, who explained how the company recently announced… Snapdragon 8 Elite Generation 5 The chip will power what could be the first phones capable of running large language models that come close to the performance of a real AI agent.

“I think we are on the cusp of the arrival of more personal assistants,” Maladi said. Referring to two LLMs, Page.AI and Anything.AIwhich held demos during the summit, saying that both could be used as personal assistants.

“It actually uses whatever local documents you have, pictures, etc., whatever is stored on the device, and just clicks on it and gets some additional information,” Malladi noted.

Organizing LLM software and appropriate hardware is part of paving the way for AI clients. But getting people to use them is another story.

Are people ready to use AI agents?

Last year, the first laptops from Qualcomm appeared Snapdragon X Elite Chips for computers first appeared. Although they don’t have a dynamically differentiated UI, they still have the basics of generative AI interfaces with prompts similar to ChatGPT. Malladi gave an example: He asked his computer about the last email he sent to Qualcomm CEO Cristiano Amon. He is now accustomed to using prompts to control his computer.

“I think people will get used to this kind of behavior, and there is no going back from there once you start getting used to it,” Maladi said.

But it will take a lot of introduction for people to understand how to use AI agents and get used to them.

“When people get more used to it, they realize how much time we spend doing a lot of mundane things — it’s a distraction from the main task you’re thinking about,” Malady said of AI.

AI agents can help people become power users of their devices without digging deep into their settings. Maladi pointed out that we often discover new features that we did not know existed in our phones, as they are such complex devices that it is no wonder that we do not know how to use them to their fullest potential. Making more capabilities easier to access by simply asking AI agents to do them could be an accessibility boon as well, Maladi said.

The vision is clear: automate mundane matters and cover up the moments when human error inadvertently creates new problems. However, generative AI has had its own struggles with accuracy, from the beginning ChatGPT including non-existent books On your psychedelic summer reading list Court cases that never happened To Google’s AI overview He ordered people to eat stones. Although Al-Mulladi confirmed that these were issues with devices with LLM certificate holders two years ago; He said guardrails have been put in place since then — both by Qualcomm at the chip level and at the LLM level.

We’ll have to wait and see if Malady is right about AI agents. The technology already exists, appearing on phones and laptops that use Qualcomm chips. If it is this year or next, the potential impact of AI agents will be significant: they will change how we use the machines that allow us to interact with and contribute to the world.

“Whether it’s school and college kids growing up in this AI environment, I wonder if they’ll know what it was like before,” Mladi mused.

There is no other way to prepare people for AI agents than to give them devices that contain the technology. After a long wait, phones may finally be able to support it, partly due to advances in Qualcomm’s mobile chipsets.

Qualcomm chips: Silicon development improves AI factors

Two years ago, the company Snapdragon 8 3rd generation It can handle 15 codes per second, which is a measure of the speed at which the AI ​​can process requests. The new Snapdragon 8 Elite Gen 5 can handle up to 220 codes per second. In short, this should allow AI assistants to process requests faster and consult more data for better responses.

“The difference between 150 codes, which was last year, to 220 codes this year, has several seconds of latency,” said Vineesh Sukumar, senior director of AI product management at Qualcomm. “Anything beyond a second impacts the human-machine interface. We think that’s not a good thing, because at some point, the user loses interest.”

In the past two years, Qualcomm has switched its mobile chip from Arm-based Kryo technology to its Oryon CPUs, boosting graphics power and neural performance. Maladi noted that the improvements made over last year’s Snapdragon 8 Elite processor – 37% better CPU performance – enable AI capabilities that are “actually a tangible difference that we will see.”

“We will always see a lot of possibilities that are available that will really help you,” Mladi said.

The company’s new Oryon CPU class debuted on the Snapdragon Trillions of operations per second – to highlight the upgrade. The new chip offers 80 TOPS, nearly double the 45 TOPS achieved by the first X Elite.

Whether these AI capabilities, combined with performance improvements over last year’s chips, will herald the era of AI agents on PC remains to be seen. Malady admits that older X Elite laptops have the same look and feel as traditional PCs despite their AI capabilities. But despite the amazing performance and battery life, ultimately, the UI hasn’t changed much. Qualcomm aims to improve the performance of the X2 Elite and Elite Extreme and increase TOPS to be enough for businesses to make the leap into turning their laptops into an AI agent interface.

In other words, hardware is no longer a barrier to creating AI customer experiences on PC and mobile devices, it is now up to manufacturers to integrate it. In these early days, there was almost no consensus about what AI agents would do. Device makers are still thinking about how to take advantage of what Qualcomm’s chips can do. A lot is still up in the air, from the number and size of LLMs that will live on the device (or connect via the cloud) to how users will actually see and interact with the AI ​​agents.

Despite this uncertainty, Mladi believes the AI ​​agent fired the gun.

“Are we waiting for something? No, probably not,” Maladi said. “The artificial agent is here, but you have to see it to believe it.” “But if you saw it, you’d say, ‘Oh yeah, I get it now.’

The first AI-powered client computer was unveiled at Snapdragon Summit

A day after Qualcomm unveiled its new mobile and PC chipsets in its keynote at Snapdragon Summit 2025, CEO Cristiano Amon sat down with Humain CEO. On a bright Maui morning, with sunlight filtering over the western mountains, the two men focused on the silver laptop between them: The first Humin computer With what they described as an AI agent interface.

That interface was more like a browser window with eight prompt templates ranging from specific (stock tip) to general (like ChatGPT “Ask Anything”). In other words, it looked more like the generative AI assistants we already know and less like the broader AI agent vision that Qualcomm has floated at this and previous summits. Qualcomm’s goal is simple: a central prompt window where people can ask questions, and an agent can search through apps and personal data to provide answers. (Disclosure: Ziff Davis, the parent company of CNET, filed a lawsuit against OpenAI in April, alleging that the company infringed Ziff Davis’s copyrights in training and operating its AI systems.)

If nothing else, it would be more training for people to start familiarizing themselves with the agent’s front-end interface rather than searching for software and answers themselves. Whether they’re typing questions on their computer keyboards or voicing them on phones, people will have to adapt to a new model of using devices.

Maladi expects that AI agents will be used differently between PC and mobile platforms.

People will ask questions on phones, and the AI ​​agent will go through one or more MBAs to get the answers. But the power limitations of a mobile device will limit its integer-based processing capabilities, mainly because it will be running on hardware. LLMs are also limited to those with relatively fewer parameters in the single digits (1, 3, 5, or 7 billion parameter models are the most common, which means lower accuracy).

On PCs, agents will focus on cloud-based models, which run from the device. These are likely to be more attractive for enterprise applications and floating-point-based models, rather than integer-based models. Eventually, they will be used to enhance productivity for work tasks, such as creating lines of code, for example.

“Half the time, you’re writing the same code you wrote before, and then you’re introducing some bugs unnecessarily,” Malady said. AI can generate a lot of that repetitive code so that programmers can focus on new approaches and problems that need to be solved.